Defamatory Political Deepfakes and the First Amendment

Case Western Reserve Law Review Case Western Reserve Law Review

Volume 70 Issue 2 Article 12

2019

Defamatory Political Deepfakes and the First Amendment Defamatory Political Deepfakes and the First Amendment

Jessica Ice

Follow this and additional works at: https://scholarlycommons.law.case.edu/caselrev

Part of the Law Commons

Recommended Citation Recommended Citation

Jessica Ice,

Defamatory Political Deepfakes and the First Amendment

, 70 Case W. Rsrv. L. Rev. 417

(2019)

Available at: https://scholarlycommons.law.case.edu/caselrev/vol70/iss2/12

This Note is brought to you for free and open access by the Student Journals at Case Western Reserve University

School of Law Scholarly Commons. It has been accepted for inclusion in Case Western Reserve Law Review by an

authorized administrator of Case Western Reserve University School of Law Scholarly Commons.

Case Western Reserve Law Review·Volume 70·Issue 2·2019

417

— Note —

Defamatory Political Deepfakes

and the First Amendment

Contents

Contents ......................................................................................... 417

Introduction .................................................................................. 417

I. What Are Deepfakes? ............................................................ 420

A. The Technology Behind Deepfakes ...................................................... 420

B. How Deepfakes Got Their Name ......................................................... 423

C. Defining Deepfakes .............................................................................. 426

II. Deepfakes of Political Figures ............................................. 427

III. Defamation Law as a Remedy ................................................. 431

IV. Satire or Parody as a Defense .............................................. 435

V. Injunctions as a Remedy ......................................................... 437

A. Injunctions and the First Amendment ................................................ 438

B. Constitutionally Permissible Injunctions on Expression ...................... 440

1. Obscenity ...................................................................................... 440

2. Copyright ...................................................................................... 441

C. Injunctions on Defamatory Speech...................................................... 442

VI. Injunctions on Defamatory Deepfakes .................................. 448

VII. Appropriate Scope for Injunctions Against Political

Defamatory Deepfakes ......................................................... 450

VIII. Injunctions and Third-Party Providers ................................ 451

Conclusion ...................................................................................... 454

Introduction

After logging onto Facebook, you see that your friend has posted a

video on your wall titled: “Donald Trump Pee Tape Released.”

Surprised, you click on the link and are directed to a YouTube video of

several young, scantily clad Russian women bouncing on a bed in what

appears to be the infamous Presidential Suite at the Ritz Carlton in

Moscow.

Donald Trump walks into the room and heads towards the

1. President Obama reportedly stayed in the Presidential Suite of the Ritz

Carlton in Moscow during a diplomatic trip to Russia. Michelle Goldberg,

Lordy, Is There a Tape?, N.Y. Times (Apr. 16, 2018), https://www.

nytimes.com/2018/04/16/opinion/comey-book-steele-dossier.html [https://

perma.cc/FB5T-GGWF]. Later, Donald Trump was alleged to have

stayed in the same room and hired Russian prostitutes to pee on the same

bed where President Obama slept. Id.

Case Western Reserve Law Review·Volume 70·Issue 2·2019

Defamatory Political Deepfakes and the First Amendment

418

women. You hear Trump’s voice say, “let’s have some fun, ladies.” The

girls laugh and you hear Trump again: “How about we show Obama

what we think of him. Why don’t you pee on his bed?” The tape cuts

off in an instant and you feel shocked. Could this video actually be real?

The video in the hypothetical above could easily be what is

described colloquially as a “deepfake.” The definition of a deepfake is

“still in flux, as technology develops.”

However, a deepfake is generally

understood to be a video made with the use of machine-learning to swap

one real person’s face onto another real person’s face. This ability

essentially makes it possible to ascribe the conduct of one individual

who has been previously videotaped to a different individual.

That is,

a deepfake is a digital impersonation of someone. This impersonation

occurs without the consent of either the person in the original video or

the person whose face is superimposed on the original. Individuals in

the public eye have already been a major target for deepfakes.

Political

figures likely will be the targets of future deepfakes, especially by those

with an interest in spreading discord and undermining public trust.

Deepfakes of political figures pose serious challenges for our political

system and even national security, but legal remedies for these videos

are complicated. Every legal remedy to combat the negative effects of

political deepfakes must go through a careful balancing test. On one

side you have a special interest in protecting high-value political speech

and furthering public discourse under the First Amendment. On the

other side you have the potential for severe public harm by undermining

elections and eroding trust in public officials. Although some political

deepfakes might be satirical and promote public discourse about the

2. James Vincent, Why We Need a Better Definition of ‘Deepfake’, The

Verge (May 22, 2018, 2:53 PM), https://www.theverge.com/2018/5/22/

17380306/deepfake-definition-ai-manipulation-fake-news [https://perma.cc/

55JA-SH67].

3. Robert Chesney & Danielle Keats Citron, Deep Fakes: A Looming

Challenge for Privacy, Democracy, and National Security, 107 Calif. L.

Rev. (forthcoming 2019) (manuscript at 4–5).

4. Samantha Cole, AI-Assisted Fake Porn Is Here and We’re All Fucked,

Vice: Motherboard (Dec. 11, 2017, 2:18 PM), https://motherboard.vice

.com/en_us/article/gydydm/gal-gadot-fake-ai-porn [https://perma.cc/

4PQW-KCVE].

5. James Vincent, US Lawmakers Say AI Deepfakes ‘Have the Potential to

Disrupt Every Facet of Our Society’, The Verge (Sept. 14, 2018, 1:17

PM),

https://www.theverge.com/2018/9/14/17859188/ai-deepfakes-national-

security-threat-lawmakers-letter-intelligence-community [https://perma.cc/

WE3U-BCLN]; Ana Romano, Jordan Peele’s Simulated Obama PSA Is a

Double-Edged Warning Against Fake News, Vox (Apr. 18, 2018, 3:00

PM), https://www.vox.com/2018/4/18/17252410/jordan-peele-obama-

deepfake-buzzfeed [https://perma.cc/3B9H-SX4X].

Case Western Reserve Law Review·Volume 70·Issue 2·2019

Defamatory Political Deepfakes and the First Amendment

419

merits of an individual candidate or issue,

many deepfakes will likely

cross over the line into pure defamation.

Given this careful balancing test, what remedies are available for

defamatory political deepfakes that survive First Amendment scrutiny?

If deepfakes are found to be truly defamatory, such speech would only

receive limited First Amendment protection and victims could recoup

monetary damages from successful defamation lawsuits.

For most

political figures, however, damages will be difficult, if not impossible,

to determine, and any monetary damages will come too late to truly

remedy the reputational harm inflicted during a campaign or their

tenure as a public figure. Thus, injunctions are likely a quicker and

more effective remedy for defamatory political deepfakes.

Although injunctions against deepfakes may seem like a logical

remedy, they will likely face major First Amendment hurdles. The

Supreme Court has yet to provide a definitive answer on whether

injunctions against defamatory speech are permissible under the First

Amendment.

Some lower courts have found injunctions to be

impermissible because they are not sufficiently tailored, effectively

creating a prior restraint on constitutionally protected speech.

Other

courts have suggested that narrowly crafted injunctions against

defamatory speech may be permissible.

Even if an injunction against

a defamatory political deepfake survives a First Amendment challenge,

victims might still be unable to remove that deepfake if its creator is

unreachable by United States courts.

This Note argues that narrowly crafted injunctions against

defamatory political deepfakes should be permitted under the First

6. See Hustler Magazine, Inc. v. Falwell, 485 U.S. 46, 54 (1988) (“Despite

their sometimes caustic nature . . . graphic depictions and satirical cartoons

have played a prominent role in public and political debate.”).

7. Ashcroft v. Free Speech Coalition, 535 U.S. 234, 245–46 (2002) (“The

freedom of speech has its limits; it does not embrace certain categories of

speech, including defamation.”).

8. See Tony v. Cochran, 544 U.S. 734 (2005) (discussing but not resolving

the permissibility of an injunction against defamatory speech).

9. See Sindi v. El-Moslimany, 896 F.3d 1, 34 (1st Cir. 2018); McCarthy v.

Fuller, 810 F.3d 456, 461–62 (7th Cir. 2015); see also infra Part V.

10. See San Antonio Cmty. Hosp. v. S. Cal. Dist. Council of Carpenters, 125

F.3d 1230, 1238 (9th Cir. 1997) (upholding an injunction against

fraudulent speech); Brown v. Petrolite Corp., 965 F.2d 38, 51 (5th Cir.

1992) (permitting a limited injunction against defamatory speech);

Lothschuetz v. Carpenter, 898 F.2d 1200, 1208 (6th Cir. 1990) (Wellford,

J., concurring in part and dissenting in part) (permitting a limited

injunction against defamatory speech); David S. Ardia, Freedom of

Speech, Defamation, and Injunctions, 55 Wm. & Mary L. Rev. 1, 41 (2013).

11. Chesney & Citron, supra note 3, at 44–45.

Case Western Reserve Law Review·Volume 70·Issue 2·2019

Defamatory Political Deepfakes and the First Amendment

420

Amendment. First, this Note gives an overview of deepfakes and the

technology used to propagate them. Then, it addresses the potential

defamatory and non-defamatory uses of political deepfakes and how

defamatory political deepfakes would be analyzed under the First

Amendment’s heightened scrutiny standard that is used to analyze

political speech. Third, it gives an overview of injunctions on speech

under First Amendment jurisprudence and provides some examples of

permissible injunctions on expression. Fourth, borrowing from obscen–

ity and copyright law, this Note discusses injunctions as a remedy for

defamatory political deepfakes and whether such injunctions should be

considered impermissible prior restraints on speech. Finally, it addresses

the issue of unreachable defendants and provides a potential solution

by extending to deepfakes the requirements for copyrighted materials

under the Digital Millennium Copyright Act.

I. What Are Deepfakes?

The notion of what a deepfake is might seem intuitive at first

glance—deepfakes seem to be a simple fake or face-swapping video. But

such a simple definition grossly oversimplifies the technology behind

deepfakes and lacks the specificity needed to properly address deepfakes

from a legal perspective.

A. The Technology Behind Deepfakes

Traditionally, any individual who wanted to edit a photo or video

would have to upload the photo or video into a computer program and

manually make any desired edits. Computer programs have gradually

made the editing process easier, but a complete manual overhaul of a

video with realistic final results is still very time- and resource-

intensive.

For instance, when the creators of Rogue One: A Star Wars

Story decided to bring back the character of Grand Moff Tarkin

through

a digital recreation, the Rogue One visual-effects supervisor described

the digital recreation process as “extremely labor-intensive and

expensive.”

12. 17 U.S.C. § 1201 (2012).

13. Donie O’Sullivan, Deepfake Videos: Inside the Pentagon’s Race Against

Disinformation, CNN (Jan. 28, 2019), https://www.cnn.com/interactive/

2019/01/business/pentagons-race-against-deepfakes/ [https://perma.cc/

PQ2H-PK59]; Scott Ross, Why VFX House Lose Money on Big Movies,

The Hollywood Reporter (Mar. 7, 2013, 5:00 AM), https://www

.hollywoodreporter.com/news/why-life-pi-titanic-vfx-426182 [https://perma

.cc/Z8KF-X2HZ].

14. Grand Moff Tarkin was originally played by the late Peter Cushing but

Cushing died prior to the filming of Rogue One: A Star Wars Story. Dave

Itzkoff, How ‘Rogue One’ Brought Back Familiar Faces, N.Y. Times

Case Western Reserve Law Review·Volume 70·Issue 2·2019

Defamatory Political Deepfakes and the First Amendment

421

Deepfakes, however, do not require human labor to manually

manipulate videos; instead, a computer’s processing power does all the

work.

The technology that makes deepfakes possible stems from “a

branch of Machine Learning that focuses on deep neural networks”

called “deep learning.”

Deep learning loosely imitates the way the

brain works by processing information through a series of nodes (similar

to neurons) in an artificial neural network. For a neural network to

replicate an image, it must take in a multitude of information from a

particular source (often called an “input layer”) and then run that

information through various nodes until it produces an “output layer.”

Neural networks are “trained” by adjusting the weights at each node

to try to improve the final “output layer” to be as close as possible to

the desired result.

Deepfakes add an extra layer of complexity onto this process

because they ultimately have two input sources: (1) the face in the

original scenario video (“original face”), and (2) the face swapped into

the original scenario video (“swapped face”). To facilitate this process,

a computer must generate two separate neural networks for each image,

each that has enough in common with the other to be able to swap

images on a shared facial structure.

A basic way to achieve this result

is through an autoencoder.

An autoencoder is a “neural network that

is trained to attempt to copy its input to its output.”

In order for the

face swapping to be successful, the computer must construct two

separate neural networks, one for the original face and another for the

swapped face, and both must be trained separately.

Once the

individual networks have been built with enough accuracy through

training, then a portion of the networks called the decoders can be

swapped, effectively pasting the swapped face onto the network of the

(Dec. 27, 2016), https://www.nytimes.com/2016/12/27/movies/how-rogue-

one-brought-back-grand-moff-tarkin.html [https://perma.cc/F9B7-8K62].

15. O’Sullivan, supra note 13.

16. Alan Zucconi, An Introduction to Neural Networks and Autoencoders,

Alan Zucconi Blog (Mar. 14, 2018), https://www.alanzucconi.com/2018/

03/14/an-introduction-to-autoencoders/ [https://perma.cc/HJ6F-MJXS].

17. Id. (explaining that image-based applications are often built on

convolutional neural networks).

18. Id.

19. Alan Zucconi, Understanding the Technology Behind DeepFakes, Alan

Zucconi Blog (Mar. 14, 2018), https://www.alanzucconi.com/2018/

03/14/understanding-the-technology-behind-deepfakes/ [https://perma.cc/

L7UE-2WWF].

20. Zucconi, supra note 16.

21. Ian Goodfellow et al., Deep Learning 499 (2016) (ebook).

22. Zucconi, supra note 19.

Case Western Reserve Law Review·Volume 70·Issue 2·2019

Defamatory Political Deepfakes and the First Amendment

422

original face. Using such technology allows the swapped face to mimic

any expressions originally made by the original face in the original

video.

Face generation can be made even more realistic through the use of

a Generative Adversarial Network (“GAN”).

Similar to autoencoding,

GANs attempt to recreate images using deep-learning techniques.

GANs achieve this result by using two components: (1) a generator,

which creates natural looking images, and (2) a discriminator, which

decides whether the images are real or fake.

Essentially the “generator

tries to fool the discriminator by generating real images as far as

possible.”

Through this adversarial process between the generator and

discriminator, the network is able to produce more consistently realistic

images than it could through a traditional autoencoding structure.

Most recently, mathematicians and computer scientists have

attempted to combine GANs with a specialized type of autoencoding

called “variational autoencoding” (collectively, “VAE-GANs”) to

23. Id.

24. Id.

25. Chesney & Citron, supra note 3, at 6; see also Ian Goodfellow et al.,

Generative Adversarial Nets, arXiv (June 10, 2014), https://arxiv.org/

pdf/1406.2661.pdf [https://perma.cc/K2HH-PL2M].

26. Chesney & Citron, supra note 3, at 7.

27. Prakash Pandey, Deep Generative Models, Medium (Jan. 23, 2018), https://

medium.com/@prakashpandey9/deep-generative-models-e0f149995b7c

[https://perma.cc/5NXU-MHC8].

28. Id.

Case Western Reserve Law Review·Volume 70·Issue 2·2019

Defamatory Political Deepfakes and the First Amendment

423

produce the most realistic output layers or generated images to date.

VAE-GANs work by using the autoencoding process to provide an

image to the GAN’s generator.

The GAN’s discriminator then checks

the image through an iterative process to make it seem more realistic.

In addition to the advancements in deepfake videos, researchers have

also made strides in improving fake audio through GANs

and other

techniques.

These advancements are only the beginning of computer-

image and audio regeneration. Deepfake creators report that the

technology behind deepfakes “is improving rapidly,” and the creators

“see no limit to whom they can impersonate.”

B. How Deepfakes Got Their Name

Although the technology (and mathematics) behind deepfakes is

complex, using deepfake technology is relatively simple.

This is

especially true after an anonymous user on reddit named “deepfake”

posted computer code on a subreddit forum in late 2017 that allowed

hobbyists to create deepfakes.

In early 2018, another reddit user

29. Enoch Kan, What The Heck Are VAE-GANs?, Towards Data Science

(Aug. 16, 2018), https://towardsdatascience.com/what-the-heck-are-vae-

gans-17b86023588a [https://perma.cc/D9J8-NURV].

30. Anders Larsen et al., Autoencoding Beyond Pixels Using a Learned Similarity

Metric, ArXiv (Feb. 10, 2016), https://arxiv.org/pdf/1512.09300.pdf

[https://perma.cc/XK4U-AQ3C].

31. Id.; Kan, supra note 29.

32. See generally Chris Donahue et al., Adversarial Audio Synthesis, ArXiv

(Feb. 12, 2018), https://arxiv.org/pdf/1802.04208.pdf

[https://perma.cc/UQ3J-892W]; Yang Gao et al., Voice Impersonation

Using Generative Adversarial Networks, ArXiv (Feb. 19, 2018),

http://arxiv.org/abs/1802.06840.pdf [https://perma.cc/FJ6K-9LC8].

33. Chesney & Citron, supra note 3, at 7; O’Sullivan, supra note 13.

34. Drew Harwell, Fake-Porn Videos Are Being Weaponized to Harass and

Humiliate Women: ‘Everybody Is a Potential Target’, Wash. Post (Dec.

30, 2018), https://www.washingtonpost.com/technology/2018/12/30/fake-

porn-videos-are-being-weaponized-harass-humiliate-women-everybody-is-

potential-target/?utm_term=.8693f1f740f5 [https://perma.cc/7TDN-662W].

35. Alan Zucconi, How to Install FakeApp, Alan Zucconi Blog (Mar. 14,

2018), https://www.alanzucconi.com/2018/03/14/how-to-install-fakeapp/

[https://perma.cc/T3G4-PEMB].

36. Jeevan Biswas, What Exactly Is Deepfakes and Why Is This AI-Based

Creation a Menace, Analytics India Mag. (Feb. 8, 2018), https://

www.analyticsindiamag.com/deepfakes-ai-celebrity-fake-videos/ [https://

perma.cc/K6E6-NJMA]. The actual month that the subreddit was created

is contested by various Internet sources. Compare id. (claiming the

“deepfakes” subreddit was created in November 2017), with Aja Romano,

Why Reddit’s Face-Swapping Celebrity Porn Craze is a Harbinger of

Dystopia, Vox (Feb. 7, 2018, 5:55 PM), https://www.vox.com/2018/1/31/

16932264/reddit-celebrity-porn-face-swapping-dystopia [https://perma.cc/

Case Western Reserve Law Review·Volume 70·Issue 2·2019

Defamatory Political Deepfakes and the First Amendment

424

adapted that code into a user-friendly application called FakeApp.

Users of the original code and FakeApp did not need to understand the

complex mathematical and computational underpinnings of deepfake

technology. Instead, users merely needed above average computer

literacy and a sufficiently powerful Graphics Processing Unit (“GPU”)

in their computer to create deepfakes.

News of the user-friendly application spread quickly, and users

began posting their own computer-generated face-swapping videos on

the “deepfake” subreddit.

User and commentators began to call the

videos “deepfakes” after the subreddit where they were born.

Many

users used the technology to perverse ends, often swapping celebrities’

faces (primarily female) onto pornographic videos.

Users also

generated “revenge porn” deepfakes by swapping their ex-girlfriends’

faces onto pornographic videos.

Not all users made pornographic

deepfakes. Many users enjoyed splicing Nicholas Cage’s face onto

various movie and television characters.

Over 80,000 people partic–

ipated in the subreddit before it was eventually shut down due to the

XP3Q-UDN7] (claiming the “deepfakes” subreddit was created in September

2017).

37. Biswas, supra note 36.

38. Zucconi, supra note 35.

39. Damon Beres & Marcus Gilmer, A Guide to ‘Deepfakes,’ the Internet’s

Latest Moral Crisis, Mashable (Feb. 2, 2018), https://mashable.com/

2018/02/02/what-are-deepfakes/#xUHFJsuHqqV [https://perma.cc/

NLP6-FBGF].

40. Julia Pimentel, Twitter and Reddit Have Banned ‘Deepfake’ Celebrity

Porn Videos, Complex (Feb. 7, 2018), https://www.complex.com/life/2018/

02/twitter-reddit-and-more-ban-deepfake-celebrity-videos [https://perma

.cc/K5TL-GHF9].

41. Cole, supra note 4; Alex Hern, AI Used to Face-Swap Hollywood Stars

into Pornography Films, The Guardian (Jan. 25, 2018), https://www

.theguardian.com/technology/2018/jan/25/ai-face-swap-pornography-emma-

watson-scarlett-johansson-taylor-swift-daisy-ridley-sophie-turner-maisie-

williams [https://perma.cc/3VDX-A5ZB].

42. Larry N. Zimmerman, Cheap and Easily Manipulated Video, 87 J. Kan.

B. Ass’n, no. 4, Apr. 2014, at 20, https://cdn.ymaws.com/www.ksbar.org/

resource/dynamic/blogs/20180410_111450_30470.pdf [https://perma.cc/

9938-ADYD].

43. Reid McCarter, Idiots on the Internet Are Getting Really Good at Splicing

Nic Cage’s Face Onto Every Movie, AV Club (Sept. 4, 2018, 10:30 AM),

https://news.avclub.com/idiots-on-the-internet-are-getting-really-good-

at-splic-1828799192 [https://perma.cc/RP9Z-QFFZ]; Sam Haysom,

Nicolas Cage is Being Added to Random Movies Using Face-Swapping

Technology, Mashable (Jan. 31, 2018), https://mashable.com/2018/01/

31/nicolas-cage-face-swapping-deepfakes/#WGhEd3yKgiqw [https://perma

.cc/SQQ3-D9M8].

Case Western Reserve Law Review·Volume 70·Issue 2·2019

Defamatory Political Deepfakes and the First Amendment

425

largely nonconsensual pornographic nature of the posted content.

Other major platforms such as Twitter, PornHub, Gyfcat, and Discord

have also banned pornographic deepfakes.

As of this writing, however,

FakeApp is still available online.

Although many commentators have described the ease of creating

deepfakes, it should be noted that downloading and using FakeApp is

probably still too cumbersome for the average computer user. First, to

“train” a neural network in a reasonable amount of time, a computer

needs an adequately sophisticated GPU.

As of 2019, a typical laptop

does not have the appropriate GPU to perform deep learning and

generate deepfakes.

In addition, although there are many tutorials on

how to download and use FakeApp,

the process requires above-

44. Beres & Gilmer, supra note 39. Later the subreddit was removed. Alex

Hern, Reddit Bans ‘Deepfakes’ Face-Swap Porn Community, The

Guardian (Feb. 8, 2018) https://www.theguardian.com/technology/2018/

feb/08/reddit-bans-deepfakes-face-swap-porn-community [https://perma.cc/

9T4L-YCDV].

45. Megan Farokhmanesh, Deepfakes Are Disappearing from Parts of the

Web, But They’re Not Going Away, The Verge (Feb. 9, 2018, 9:00 AM),

https://www.theverge.com/2018/2/9/16986602/deepfakes-banned-

reddit-ai-faceswap-porn [https://perma.cc/C4E5-D2PH].

46. FakeApp Download, https://www.malavida.com/en/soft/fakeapp/#gref

[https://perma.cc/P87C-SEGP] (last visited Nov. 13, 2018).

47. Zucconi, supra note 35; Slav Ivanov, Picking a GPU for Deep Learning,

Slav Ivanov Blog (Nov. 22, 2017), https://blog.slavv.com/picking-a-

gpu-for-deep-learning-3d4795c273b9 [https://perma.cc/UZZ5-SX66].

48. See Nicholas Deleon, Why Even Non-Gamers May Want a Powerful

Graphics Card in Their Next Computer, Consumer Reports (Aug. 29,

2018), https://www.consumerreports.org/computers/why-even-non-gamers-

may-want-a-powerful-graphics-card/ [https://perma.cc/JVS7-7DA3]

(indicating that consumers need to elect to upgrade their laptops to include

GPUs at the time of purchase); Janakiram MSV, In the Era of Artificial

Intelligence, GPUs Are the New CPUs, Forbes (Aug. 7, 2017, 10:14AM),

https://www.forbes.com/sites/janakirammsv/2017/08/07/in-the-era-of-

artificial-intelligence-gpus-are-the-new-cpus/#6e5d76755d16 [https://perma

.cc/TU2T-J49H] (noting that for the average consumer GPUs were purely

optional); Zucconi, supra note 35 (stating the training the neural network

without a GPU would take weeks instead of hours).

49. See generally Zucconi, supra note 35; tech 4tress, Deepfakes Guide: Fake

App 2 2 Tutorial. Installation (Totally Simplified, Model Folder Included),

YouTube (Feb. 21, 2018), https://www.youtube.com/watch?v=Lsv38PkLsGU&

t=83s [https://perma.cc/FVL3-23WU]; Irrelevant Voice, Deepfakes

Tutorial (FakeApp) (Fake Adult Videos of Celebrities), YouTube (Jan.

28, 2018), https://www.youtube.com/watch?v=ghTb2kZSpZE [https://

perma.cc/YV6C-5PGQ]; Oliver Lardner, Tutorial for Mac: Deepfakes—

Reddit [MIRROR], Medium (Feb. 7, 2018), https://medium.com/@

oliverlardner/tutorial-for-mac-deepfakes-reddit-mirror-d75eb8069a16 [https://

perma.cc/ER5V-T9PW].

Case Western Reserve Law Review·Volume 70·Issue 2·2019

Defamatory Political Deepfakes and the First Amendment

426

average computer literacy, including understanding torrenting, path

configuration, file structures, and application versioning.

Although the

GPU and technological skills necessary to create deepfakes may be

hurdles for the average computer user, they are far less cumbersome for

an avid computer-hobbyist or gamer. As of 2018, individuals could buy

GPUs sufficient to create deepfakes for as low as $160.

Although

regular laptops may be inadequate for creating deepfakes, gaming

laptops regularly feature sufficiently powerful GPUs.

As developers

create more user-friendly deepfake applications, average computer users

will likely gain greater deepfake-creating capabilities. But even if the

process of creating deepfakes becomes easier, it will likely always

involve deliberate affirmative actions on behalf of the creator (such as

selecting which faces or scenarios to swap).

C. Defining Deepfakes

As noted above, the technology driving deepfakes and computer-

generated images is still rapidly evolving. Thus, defining deepfakes is

especially tricky and researchers are still struggling with a uniform

definition.

Legal scholars Robert Chesney and Danielle Citron

describe deepfake technology as “leverag[ing] machine-learning algor–

ithms to insert faces and voices into video and audio recordings of

actual people and enabl[ing] the creation of realistic impersonations out

of digital whole cloth.”

Building on this definition, legal scholar

Richard Hasen defined deepfakes as “audio and video clips manipulated

using machine learning and artificial intelligence that can make a

politician, celebrity, or anyone else appear to say or do anything the

manipulator wants.”

Artificial intelligence researcher Miles Brundage

noted that the term “deepfake” generally refers to a “subset of fake

video that leverages deep learning . . . to make the faking process

easier.”

Technologist Aviv Ovadya said that deepfakes can be

50. Zucconi, supra note 35.

51. Ivanov, supra note 47.

52. Gordon Mah Ung & Alaina Yee, How to Pick the Best Gaming Laptop

GPU, PCWorld (Sept. 5, 2018, 4:53 PM), https://www.pcworld.com/

article/3237991/laptop-computers/how-to-pick-the-best-gaming-laptop-

gpu.html [https://perma.cc/ES33-UZX2].

53. Vincent, supra note 2.

54. Chesney & Citron, supra note 3, at 4.

55. Richard L. Hasen, Deep Fakes, Bots, and Siloed Justices: American

Election Law in a Post-Truth World, St. Louis U. L. J., 9 (forthcoming

2020). This Note quotes and cites this article with Professor Hasen’s

permission.

56. Vincent, supra note 2; Miles Brundage et. al., The Malicious Use of

Artificial Intelligence: Forecasting, Prevention, and Mitigation, Future

of Human. Inst. (Feb. 2018), https://img1.wsimg.com/blobby/go/

Case Western Reserve Law Review·Volume 70·Issue 2·2019

Defamatory Political Deepfakes and the First Amendment

427

described as “audio or video fabrication or manipulation that would

have been extremely difficult and expensive without AI advances.”

All of these definitions include the method of creation (by deep

learning or artificial intelligence) as a key way to distinguish deepfakes

from other faked videos. This distinction is very important because the

use of deep learning in a video’s creation implies that such a video can

be created more easily than a manually manipulated video. In addition,

as the technology improves, videos created by deep learning have the

potential to look more realistic than manually altered videos. Another

important element of deepfake videos is the “faking” of a face or a

scenario in a pre-existing video or image. Although to date most

deepfakes have primarily focused on face-swapping and voice alter–

ations, the technology could be used to swap out other components in

a video, such as background or objects in a video. A final definition

must be flexible enough to incorporate any “faking” of an original video

possible by deep learning technology. Combining all the key elements

above, we can use this simplified definition: Deepfakes are videos,

images, and audio created using deep learning to alter the content of

an original video, image, and/or audio file by face-swapping or scenario

alterations.

II. Deepfakes of Political Figures

As we have seen in Part I, deepfakes have a wide range of uses and

raise many legal issues. This Note focuses solely on the impact of

deepfakes targeting political figures, or “political deepfakes.” Through–

out history, individuals have been inclined to use the latest

communications technology to mock, comment about, and criticize

politicians.

Persons in power have been suspicious of such technology,

often banning or severely restricting its dissemination, usually at the

expense of growth and the exchange of ideas.

For instance, upon the

invention of the printing press in 1456, French King Charles VII sent a

spy to Mainz to investigate how the device might be used to spread

political ideas.

Charles was concerned that the technology could

disseminate information very quickly; he was relieved once he learned

that government leaders placed strict controls on what information

3d82daa4-97fe-4096-9c6b-376b92c619de/downloads/1c6q2kc4v_50335.pdf

[https://perma.cc/ZRA2-69RE].

57. Vincent, supra note 2.

58. Craig Smith et al., The First Amendment—Its Current Condition, in The

First Amendment—The Challenge of New Technology 9–12 (Sig

Mickelson & Elena Mier Y. Teran eds., 1989).

59. Id.

60. Id. at 9–10.

Case Western Reserve Law Review·Volume 70·Issue 2·2019

Defamatory Political Deepfakes and the First Amendment

428

could be published.

History teaches us that if we choose to attempt to

mitigate the risks of a new communication technology, such restrictions

should be as limited as possible so as not to curtail the benefits of that

technology.

From the printing press to the Internet, deepfakes are simply the

next in a long line of disruptive communications technology that can

be used to further civic discourse, or misused to deceive and undermine

public trust. Individuals have already used deepfakes to further public

discussion. Jordan Peele created a political deepfake of President

Obama to warn the public of the dangers of political deepfakes in April

of 2018.

Many political deepfakes, such as Jordan Peele’s, will likely

be viewed by courts as “speech on public issues,” which deserves a place

on the “‘highest rung of the hierarchy of First Amendment values’ and

is entitled to special protection.”

Given the possible beneficial uses of

deepfakes by satirists, educators, and artists, an outright ban on

deepfake technology is not only ill-advised but also likely uncon–

stitutional.

However, because deepfakes use deep learning to reduce the effort

needed to “fake” a previously made video, bad actors will have a special

incentive to use this technology not only to mock, but potentially to

frame, undermine, or blackmail political figures. Chesney and Citron

outlined eight potential harms to society resulting from the use of

deepfakes: (1) distortion of democratic discourse, (2) manipulation of

elections, (3) eroding trust in institutions, (4) exacerbating social

divisions, (5) undermining public safety, (6) undermining diplomacy,

(7) jeopardizing national security, and (8) undermining journalism.

Political deepfakes are the most likely candidates to perpetrate almost

all of the harms outlined by Chesney and Citron. For instance, if a

political deepfake depicting a candidate taking a bribe gets released on

the eve of an election, such a tape could simultaneously change the

election results, distort discourse about the candidates, and erode trust

in public figures and institutions. If the tape agitates activist groups

enough that they begin to hold public protests or demonstrations, such

demonstrations could turn violent and undermine public safety. If a

news outlet runs the story believing it is real and only later finds out it

is fake, the deepfake could also undermine the public’s faith in

61. Id.

62. David Mack, This PSA About Fake News from Barack Obama Is Not What

it Appears, BuzzFeed News (Apr. 17, 2018), https://www.buzzfeednews

.com/article/davidmack/obama-fake-news-jordan-peele-psa-video-buzzfeed#

.el7Eqkeo7A [https://perma.cc/PDW2-DQSV].

63. Snyder v. Phelps, 562 U.S. 443, 444 (2011) (quoting Connick v. Myers,

461 U.S. 138, 145 (1983)).

64. Chesney & Citron, supra note 3, at 20–28.

Case Western Reserve Law Review·Volume 70·Issue 2·2019

Defamatory Political Deepfakes and the First Amendment

429

journalism. Although this doomsday scenario may seem extreme,

deepfakes that manifest even one of these threats to society could still

cause severe harm.

The deepfake threat has not gone unnoticed by Congress. In July

2018, Senator Marco Rubio mused on a potential danger of deepfakes

as “the ability to influence the outcome [of an election] by putting out

a video of a candidate on the eve before the election doing or saying

something strategically placed, strategically altered, in such a way to

drive some narrative that could flip enough votes in the right place to

cost someone an election.”

Senator Rubio characterized such a

situation as “not a threat to our elections, but a threat to our Repub–

lic.”

Senator Mark Warner also specifically addressed deepfakes in his

draft white paper outlining various policy proposals to regulate social

media and large technology firms.

On September 13, 2018,

Representatives Adam Schiff, Stephanie Murphy, and Carlos Curbelo

sent a letter to Director of National Intelligence Dan Coates specifically

asking the intelligence community to “report to Congress and the public

about the implications of new technologies that allow malicious actors

to fabricate audio, video, and still images.”

Some legislators have even gone so far as to introduce specific

legislation to address deepfakes.

On December 21, 2018, Senator Sasse

introduced a bill attempting to criminalize the “malicious creation and

distribution of deepfakes” entitled the “Malicious Deep Fake

Prohibition Act of 2018.”

The bill makes it a crime punishable by up

to ten years imprisonment to:

65. Senator Marco Rubio, Keynote Remarks at The Heritage Foundation’s

Homeland Security Event on Deep Fakes (July 19, 2018), video available

at Deep Fakes: A Looming Challenge for Privacy, Democracy, and National

Security, The Heritage Foundation, at 15:00–15:21, https://www.

heritage.org/homeland-security/event/deep-fakes-looming-challenge-privacy-

democracy-and-national-security [https://perma.cc/G97A-HK5J].

66. Id.

67. Senator Mark R. Warner, Potential Policy Proposals for Regulation of

Social Media and Technology Firms 2 (Aug. 20, 2018) (White Paper

Draft), https://www.ftc.gov/system/files/documents/public_comments/

2018/08/ftc-2018-0048-d-0104-155263.pdf [https://perma.cc/8Y5F-DUQP].

68. Letter from Adam B. Schiff, Stephanie Murphy, and Carlos Curbelo,

Representatives, U.S. House of Representatives, to Daniel R. Coats, Dir.,

Office of Nat’l Intelligence (Sept. 13, 2018), https://schiff.house.gov/imo/

media/doc/2018-09%20ODNI%20Deep%20Fakes%20letter.pdf [https://

perma.cc/D82S-TPQV].

69. Kaveh Waddell, Lawmakers Plunge into “Deepfake” War, Axios (Jan.

31, 2019), https://www.axios.com/deepfake-laws-fb5de200-1bfe-4aaf-9c93-

19c0ba16d744.html [https://perma.cc/JCP3-9HDF].

70. Id.; S. 3805, 115th Cong. § 2 (2018). The bill defines “deep fake” as “an

audiovisual record created or altered in a manner that the record would

Case Western Reserve Law Review·Volume 70·Issue 2·2019

Defamatory Political Deepfakes and the First Amendment

430

(1) create, with the intent to distribute, a deep fake with the

intent that the distribution of the deep fake would facilitate

criminal or tortious conduct under Federal, State, local, or Tribal

law; or

(2) distribute an audiovisual record with –

(A) actual knowledge that the audiovisual record is a deepfake;

and

(B) the intent that the distribution of the audiovisual record

would facilitate criminal or tortious conduct under Federal, State,

local or Tribal law.

The bill includes the limitation that “any activity protected by the

First Amendment” is not punishable under the Act.

Although the bill

expired at the end of 2018, Sasse’s office indicated that it intends to

reintroduce the bill.

Representative Clarke of New York introduced a

bill entitled the “DEEP FAKES Accountability Act in June of 2019.”

This bill requires deepfake creators to include either a watermark or

disclosure that the audio or video is altered.

Deepfake creators who

do not comply with the watermark or disclosure requirements may face

civil penalties of up to $150,000 per record in fines, five years in jail, or

both.

The bill also allows for injunctive relief to comply creators to

use watermarks or disclosures.

In addition to the legislation proposed

by Congress, several states, including New York, Texas, and Maine,

introduced legislation to address the harms of deepfakes.

In October

falsely appear to a reasonable observer to be an authentic record of the

actual speech or conduct of an individual.” Id. Strikingly, the bill’s

definition of deepfake does not restrict the manner of creation to videos

created through deep learning. See id.; supra Part I.C.

71. S. 3805, 115th Cong. § 2 (2018). Law professor and commentator Orin

Kerr noted his concern that Senator Sasse’s bill reaches too far, creating

“federal crimes that prohibit acts undertaken in furtherance of any

criminal law or tort.” Orin Kerr, Should Congress Pass a “Deep Fakes”

Law?, The Volokh Conspiracy (Jan. 31, 2019), http://reason.com/

volokh/2019/01/31/should-congress-pass-a-deep-fakes-law [https://perma

.cc/43DG-9WBU].

72. S. 3805, 115th Cong. § 2 (2018).

73. Waddell, supra note 69.

74. H.R. 3230, 116th Cong. § 1 (2019).

75. Id. § 2.

76. Id.

77. Id.

78. Nina Iacono Brown, Congress Wants to Solve Deepfakes by 2020, Slate

(July 15, 2019, 7:30A.M.), https://slate.com/technology/2019/07/congress-

Case Western Reserve Law Review·Volume 70·Issue 2·2019

Defamatory Political Deepfakes and the First Amendment

431

of 2019, California passed legislation penalizing the distribution of

deepfakes of political candidates within 60 days of an election.

Although both federal and state level legislatures have attempted to

address the harms of deepfakes, whether these measures will comport

with the First Amendment has yet to be tested in court.

Given all of these concerns, this Note must return to our original

hypothetical. Imagine that you believe that the Donald Trump’s Pee

Tape video posted by your friend is real. Imagine that you are not the

only one. Thousands of people believe that this video is the smoking

gun that proves both that the Russians had information that could have

compromised Trump and that Russia largely controlled him throughout

the 2016 election and into his presidency. The video contains imagery

so graphic that even the Republican base turn against Trump.

Democratic activist groups are primed to believe such a video and take

to the streets, protesting in outrage. The House votes to impeach

Trump on newly discovered evidence of corruption. Republican

Senators, fearing the repercussions of not acting, ultimately decide to

remove Trump from office. Only after all of this has taken place does

video forensics reveal that the video was a deepfake. In this scenario,

could Trump recover damages from the video’s creator? What would

those damages be?

III. Defamation Law as a Remedy

For defamatory deepfakes in general, individuals or companies may

have legal remedies to protect their property or publicity under

However, politicians

will likely struggle to recover under either of these legal theories.

Regarding copyright law, political figures often do not own much of the

video taken of them, barring them from any recovery.

Even if a

political figure owns the video of themselves, courts will likely not

punish an individual who uses the video to comment on the political

deepfake-regulation-230-2020.html [https://perma.cc/HT5S-HBVN]; Scott

Thistle, Maine Joins ‘Deepfake’ Flight: Might Ban in Political Ads,

Governing (Jan. 30, 2020), https://www.governing.com/next/Maine-

Joins-Deepfake-Fight-Might-Ban-in-Political-Ads.html [https://perma.cc/

3CFY-WBM4].

79. Colin Lecher, California has Banned Political Deepfakes During Election

Season, Verge (Oct. 7, 2019), https://www.theverge.com/2019/10/7/

20902884/california-deepfake-political-ban-election-2020 [https://perma.cc/

88Y7-4AA2].

80. Chesney & Citron, supra note 3, at 34–35.

81. Hustler Magazine, Inc. v. Moral Majority, Inc., 796 F.2d 1148, 1151 (9th

Cir. 1986) (noting the first element of copyright infringement is

“ownership of the copyright by the plaintiff”).

Case Western Reserve Law Review·Volume 70·Issue 2·2019

Defamatory Political Deepfakes and the First Amendment

432

figure in a non-commercial way, categorizing such commentary as “fair

use.” For instance, in Dhillon v. Doe,

the United States District Court

for the Northern District of California held that a non-commercial

website’s use of a politician’s headshot to criticize the politician was

“transformative” and “such a use is precisely what the Copyright Act

envisions as a paradigmatic fair use.”

Public figures will likely also

struggle to recover under the right of publicity, or “the inherent right

of every human being to control the commercial use of his or her

identity.”

First, only about thirty states even recognize such a right.

Second, and even more importantly, the right to publicity only protects

the use of someone’s persona for a commercial gain.

As in the Trump

hypothetical, many political deepfakes do not seek to profit from their

creations, so the right of publicity will not provide an appropriate legal

remedy.

This brings us to the pièce de résistance: the tort of defamation.

Defamation law covers both libel (written communications or commun–

ications that persist similarly to written works) and slander (spoken

communications).

Although the actual elements of defamation vary

from state to state, the tort generally consists of several elements as

outlined in the Second Restatement of Torts:

To create liability for defamation there must be:

(a) a false and defamatory statement concerning another;

(b) an unprivileged publication to a third party;

publisher; and

(d) either actionability of the statement irrespective of special

harm or the existence of special harm caused by the publication.

82. No. C 13-01465 SI, 2014 WL 722592 (N.D. Cal. Feb. 25, 2014).

83. Id. at *5.

84. 1 J. Thomas McCarthy, Rights of Publicity and Privacy § 3:1

(2d ed. 2018).

85. Amanda Tate, Note, Miley Cyrus and the Attack of the Drones: The Right

of Publicity and Tabloid Use of Unmanned Aircraft Systems, 17 Tex.

Rev. Ent. & Sports L. 73, 84 (2015).

86. McCarthy, supra note 84.

87. Robert D. Sack, Sack on Defamation: Libel, Slander, and

Related Problems §§ 2:4.1–.2 (4th ed. 2010).

88. Restatement (Second) of Torts § 558 (Am. Law Inst. 1977).

Case Western Reserve Law Review·Volume 70·Issue 2·2019

Defamatory Political Deepfakes and the First Amendment

433

However, in New York Times Co. v. Sullivan,

the Supreme Court

replaced the traditional third element of “fault amounting to at least

negligence” with a higher standard, requiring proof that the speech’s

publisher acted with actual malice in publishing speech directed at

public officials.

Actual malice is found when the statement is made

“with knowledge that it was false or with reckless disregard of whether

it was false or not.”

Mere negligence does not suffice to meet an actual-

malice standard.

The actual-malice standard creates an exceptionally high burden of

proof for public officials in defamation suits, as it is very difficult to

prove that a publisher knew the information she published was false or

that her actions regarding the information’s truthfulness rose to the

level of recklessness. The Supreme Court created such a high burden to

balance the competing interests of public officials who sought to protect

their reputations from defamation against the First Amendment’s

safeguarding of high-value political speech. The Court reasoned that

erroneous statements are “inevitable in free debate,” adding that even

erroneous statements should be protected to provide the freedom-of-

expression principle the adequate “breathing space” it needs to

survive.

The Court held that when public officials bring a defamation

suit, any mental state on behalf of the publisher lower than actual

malice is “constitutionally deficient [in its] failure to provide the

safeguards for freedom of speech and of the press that are required by

the First and Fourteenth Amendments.”

Such a standard still

provides recourse for political figures because if they can prove that

individuals acted with actual malice when publishing false and defam–

atory statements about them, then the statement is not protected under

the First Amendment.

Applying this rigorous standard to political deepfakes is relatively

straightforward. Deepfakes are intended to be “fakes” or falsehoods.

Unlike an article where an individual might not be sure about one

particular fact regarding a political figure, creators of deepfakes know

89. 376 U.S. 254 (1964).

90. Id. at 283; Restatement (Second) of Torts § 558 (Am. Law Inst.

1977). The Court later extended the holding from Sullivan to apply not

only to public officials but also to public figures. Curtis Publishing Co. v.

Butts, 388 U.S. 130, 150 (1967).

91. Sullivan, 376 U.S. at 280.

92. Id. at 287–88.

93. Id. at 271–72.

94. Id. at 264.

95. See Ashcroft v. Free Speech Coalition, 535 U.S. 234, 245–46 (2002) (“The

freedom of speech has its limits; it does not embrace certain categories of

speech, including defamation . . . .”).

Case Western Reserve Law Review·Volume 70·Issue 2·2019

Defamatory Political Deepfakes and the First Amendment

434

that the content they are creating is false. Creating a deepfake requires

a person with above-average computer literacy to download a program

and take a series of non-trivial, affirmative steps to swap someone’s face

onto another video.

Even as technology makes it easier to create

deepfakes, it will still be hard to prove that an individual “accidentally”

created a video of a public figure in a compromising situation. Deep–

fakes are designed to look realistic and to make others believe that the

face swapped onto the fake video is in fact the person it seems to be.

The old adage “seeing is believing” still holds as videos are generally

considered highly reliable evidence of past events.

A deepfake creator

will not only know that what she is creating is false, but since she is

creating a video, many more persons are likely to believe that what the

video depicts is true.

Nevertheless, good-faith deepfake creators will have a rather easy

shield against culpability: any indication, either in the video itself or on

the location (webpage) where the video was posted, that the video is a

fake. If a deepfake creator supplies a watermark or disclaimer, clearly

indicating the video is a fake, such information could be used as

compelling evidence that the creator reasonably believed that others

would not treat the video as an actual representation of the political

figure. If a third party could not reasonably believe the video was a

depiction of the political figure, then the video is not defamatory.

Given the ease of providing such a disclaimer or watermark, this will

also work against bad actors who intend to use the video to deceive. If

an individual creates a defamatory deepfake without providing any

indication that the video is fake, it seems likely that she knew or had

reckless disregard for the video’s falsity.

In addition to the actual-malice standard, politicians will also have

to establish the other elements of defamation: “a false and defamatory

statement concerning another,” “an unprivileged publication to a third

party,” and “either actionability of the statement irrespective of special

harm or the existence of special harm caused by the publication.”

The

Second Restatement of Torts defines a defamatory statement as one

that “tends so to harm the reputation of another as to lower him in the

estimation of the community or to deter third persons from associating

or dealing with him.”

100

Some videos will be completely innocuous (e.g.,

96. See supra Part I.B.

97. Brundage et. al., supra note 56, at 46.

98. See Hustler Magazine, Inc. v. Falwell, 485 U.S. 46, 57 (1988) (holding

that the publisher was not liable for defamation because the parody ad

“could not be reasonably understood as describing actual facts about

[respondent]”).

99. Restatement (Second) of Torts § 558 (Am. Law Inst. 1977).

100. Id. § 559.

Case Western Reserve Law Review·Volume 70·Issue 2·2019

Defamatory Political Deepfakes and the First Amendment

435

a video of a politician watering her garden) and thus, not defamatory.

Bad actors, however, are likely to be more interested in using deepfake

technology to place political figures in compromising situations, aiming

to lower the reputation of those political figures. The trier of fact will

have the final word on whether the content is deemed defamatory.

Political figures seeking to recover damages will also have to prove

that the deepfake was published and that they were harmed by that

publication. A statement is published through “any act by which the

defamatory matter is . . . communicated to a third person.”

101

Posting

a deepfake online would certainly constitute a publication. Finally, the

political figure could argue that the publication caused her special harm

(or some kind of monetary loss), such as the loss of an election or

reduced campaign donations.

102

Even if the political figure did not suffer

a particularized harm, she could argue that the publication caused

general societal harm (such as the harms outlined in Part II).

IV. Satire or Parody as a Defense

A good-faith deepfake creator’s most readily available defense is an

acknowledgement in or near the video that the video is false. The more

obvious the disclaimer or watermark, the less likely such a video will

be misconstrued by a member of the general public that such a video

is actually real. Some political satirists, however, might not want to be

constrained by the addition of a watermark or disclaimer and instead

may claim a privilege by satire or parody. The Supreme Court noted

the importance of satire in the political-discourse context in Hustler

Magazine, Inc. v. Falwell, stating that “[d]espite their sometimes

caustic nature . . . graphic depictions and satirical cartoons have played

a prominent role in public and political debate.”

103

In that case, Jerry Falwell sued Hustler Magazine for defamation

and intentional infliction of emotional distress after Hustler editors

drew a cartoon in which Falwell discussed his “first time” as a “drunken

incestuous rendezvous with his mother.”

104

At trial, the jury rejected

Falwell’s defamation claim based on its findings that no reasonable

person would believe that the cartoon described actual facts; however,

it found in favor of Falwell on his intentional infliction of emotion

distress claim.

105

On appeal, Hustler Magazine claimed that awarding

damages for intentional infliction of emotional distress violated its First

101. Id. § 577 cmt. a.

102. Ardia, supra note 10, at 9.

103. 485 U.S. 46, 54 (1988).

104. Id. at 48.

105. Id. at 49.

Case Western Reserve Law Review·Volume 70·Issue 2·2019

Defamatory Political Deepfakes and the First Amendment

436

Amendment rights.

106

Falwell contended that the cartoon was so

“outrageous” that it was distinguishable from a traditional political

cartoon.

107

The Court, inherently suspicious of an “outrageousness”

argument, rejected Falwell’s position, noting:

“Outrageousness” in the area of political and social discourse has

an inherent subjectiveness about it which would allow a jury to

impose liability on the basis of the jurors’ tastes or views, or

perhaps on the basis of their dislike of a particular expression. An

“outrageousness” standard thus runs afoul of our longstanding

refusal to allow damages to be awarded because the speech in

question may have an adverse emotional impact on the

audience.

108

The Court instead held that “public figures . . . may not recover

for the tort of intentional infliction of emotional distress by reason of

[satirical] publications . . . without showing in addition that the

publication contains a false statement of fact which was made with

‘actual malice.’”

109

The Court adopted New York Times’s actual-malice

standard to the tort of intentional infliction of emotional distress to

provide “adequate ‘breathing space’ to the freedoms protected by the

First Amendment.”

110

The Supreme Court’s view of outrageousness might seem like a

blank check for satirists and parody makers to use deepfakes to mock

politicians. The question remains, however, whether deepfakes can be

considered satirical works in general. The satire at issue in Hustler was

a hand-drawn caricature that the jury found could not “reasonably be

understood as describing actual facts about [respondent] or actual

events in which [he] participated.”

111

Deepfakes, unlike caricatures and

other parodies, could reasonably be understood to depict facts about an

individual. This reasonableness element is crucial for establishing

liability against a defendant asserting a parody or satire defense. In

Farah v. Esquire,

112

the United States Court of Appeals for the District

of Columbia Circuit noted that “satire is effective as social commentary

precisely because it is often grounded in truth.”

113

Thus, “[t]he test [for

satire] . . . is not whether some actual readers were misled, but whether

106. Id. at 50.

107. Id. at 55.

108. Id.

109. Id. at 56.

110. Id.

111. Id. at 49.

112. 736 F.3d 528 (D.C. Cir. 2013).

113. Id. at 537.

Case Western Reserve Law Review·Volume 70·Issue 2·2019

Defamatory Political Deepfakes and the First Amendment

437

the hypothetical reasonable reader could be (after time for

reflection).”

114

What it takes to mislead the “hypothetical reasonable

reader” will certainly evolve as deepfake technology improves to mirror

true human interaction. If the video is extremely realistic, and the

actions are outrageous but within the realm of possibility, a court may

find that a deepfake intended to be a satire is actually defamatory

speech because a reasonable viewer would not be able to distinguish the

fake video from reality.

V. Injunctions as a Remedy

If a political figure can prove that a publisher acted with actual

malice to create a defamatory deepfake, and that the deepfake could

not be considered a satire, then the political figure will be able to

recover the traditional tort remedy of monetary damages.

115

Despite the

attraction of monetary damages, they do nothing to stop the ongoing

reputational loss caused by the deepfake’s continued existence on the

Internet.

116

Nor do monetary damages remedy the societal damages

caused by the video if it is allowed to persist in the public sphere.

Moreover, many bad actors posting deepfakes might be judgment-proof,

denying the political figure any recovery.

Given the inadequacy of monetary damages, injunctions may be a

better avenue to address some of the harms caused by deepfakes. Unlike

monetary damages, an injunction is an equity-based remedy designed

to “order the defendant to refrain from specified conduct.”

117

Political

figures are likely to favor injunctions because removing the original

video from the Internet will remedy the imminent issue of reputational

loss caused by a defamatory political deepfake.

118

Courts can grant

permanent injunctions that occur after an adjudication on the merits

of the claim, or preliminary injunctions that occur prior to adjudications

based on the likelihood of success on the merits.

119

Although preliminary

injunctions may be favored because they remove an offending video

from the Internet sooner than other remedies, they face a heightened

First Amendment hurdle because they risk censoring protected

114. Id.

115. Ardia, supra note 10, at 7.

116. Id. at 16.

117. Id. at 14.

118. Id. at 9, 15–16.

119. Mark A. Lemley & Eugene Volokh, Freedom of Speech and Injunctions

in Intellectual Property Cases, 48 Duke L.J. 147, 149–50 (1998).

Case Western Reserve Law Review·Volume 70·Issue 2·2019

Defamatory Political Deepfakes and the First Amendment

438

speech.

120

Thus, this Note focuses primarily on the potential use of

permanent injunctions against defamatory political deepfakes.

A. Injunctions and the First Amendment

The First Amendment protects individuals’ rights to freedom of

religion, speech, the press, and assembly.

121

Since the founding of the

United States, courts have been keen to avoid prior restraints on speech,

or restraints on expression before occurs.

122

In Near v. Minnesota,

123

the

Supreme Court established that injunctions against free speech and the

press usually amount to a prior restraint on expression.

124

In Near, the

Court held a state statute that perpetually enjoined individuals or

corporations (including the press) who published “a malicious,

scandalous and defamatory newspaper or other periodical” was an

unconstitutional prior restraint on freedom of the press.

125

The broad

nature of the statute led to the “effective censorship” of any publisher

who was convicted of publishing an “offending newspaper or

periodical,” because all future publications regarding the offending

matter (defamatory or not) would be subject to court review.

126

The

Court held that the threat of enjoining the press’ constitutionally

protected activity outweighed the potential that publishing scandalous

and defamatory information may disturb the public peace.

127

Similarly, the Supreme Court struck down injunctions against press

activities in both New York Times Co. v. United States (Pentagon

Papers)

128

and Nebraska Press Association v. Stuart

129

as uncon–

stitutional prior restraints on the freedom of speech and the press. In

Pentagon Papers, the Second Circuit enjoined the New York Times and

the Washington Post from publishing the “Pentagon Papers,” a Defense

Department study of American activities in Southeast Asia, after the

government argued that the information’s publication would endanger

120. See id. at 150.

121. U.S. Const. amend. I.

122. See Anthony Lewis, Make No Law: The Sullivan Case and the

First Amendment 51 (1991); Ardia, supra note 10, at 32. The perception

that injunctions were prior restraints on speech and therefore invalid as

an equitable remedy was so pervasive commentators described it as the

“no injunction rule.” Id. at 18.

123. 283 U.S. 697 (1931).

124. Id. at 722–23.

125. Id. at 702–03, 722–23.

126. Id. at 712.

127. Id. at 721–22.

128. (Pentagon Papers) 403 U.S. 713 (1971) (per curiam).

129. 427 U.S. 539 (1976).

Case Western Reserve Law Review·Volume 70·Issue 2·2019

Defamatory Political Deepfakes and the First Amendment

439

national security.

130

In Nebraska Press, a state court feared that

reporting on a sensational criminal trial in a small community would

endanger the defendant’s right to a fair trial.

131

The court crafted a

restrictive order prohibiting the press from reporting on certain aspects

of the trial.

132

In both cases, the Supreme Court grappled with a careful

balancing test between constitutionally protected speech and important

governmental interests (national security) or individual rights (the right

to a fair trial).

133

And in both cases, the Supreme Court held that the

First Amendment’s free-speech guarantee outweighed either competing

interest.

134

In the case of deepfakes, courts must also balance the speech

contained in defamatory political deepfakes against important

governmental interests, such as safeguarding free elections, promoting

trust in public officials, defending national security, and protecting

truth itself.

135

It is important to note that in Near, Pentagon Papers,

and Nebraska Press, the balancing test involved either a statute or a

court-ordered remedy that both targeted an entire industry (the press)

and included speech that likely was not libelous.

136

None of the cases

addressed an individual or private wrong of one person defaming

another. For political deepfakes, it is likely the defamatory threat would

not come from the press, but instead from a private individual

intentionally attempting to disparage a political figure. In addition,

unlike the cases discussed above, the expression in the deepfake would

be clearly false.

137

Thus, if an injunction could be carefully crafted to

include only the speech in the defamatory deepfake itself, the resulting

balancing-test analysis would pit unprotected false and defamatory

speech against important governmental interests. To determine

whether a court might hold that the important (some might even say

“compelling”) governmental interests outweigh any free-speech con–

cerns, we must take a closer look at several constitutionally permissible

prior restraints on expression.

130. Pentagon Papers, 403 U.S. at 714.

131. Nebraska Press, 427 U.S. at 543–45.

132. Id. at 543–44.

133. Pentagon Papers, 403 U.S. at 718; Nebraska Press, 427 U.S. at 543.

134. Pentagon Papers, 403 U.S. at 714; Nebraska Press, 427 U.S. at 570.

135. See supra Part II.

136. Near v. Minnesota, 283 U.S. 697, 702–03 (1931); Pentagon Papers, 403

U.S. at 714; Nebraska Press, 427 U.S. at 570.

137. See supra Part I.

Case Western Reserve Law Review·Volume 70·Issue 2·2019

Defamatory Political Deepfakes and the First Amendment

440

B. Constitutionally Permissible Injunctions on Expression

Although prior restraints on expression bear “a heavy presumption

against [their] constitutional validity,” not all prior restraints are pro–

hibited by the Constitution.

138

In Near, the Supreme Court outlined

four exceptions to the prior-restraint rule: (1) key issues of national

security, such as troop movements or sailing dates; (2) obscene

publications; (3) incitement to acts of violence or attempts to overthrow

the government; and (4) to “protect private rights” in accordance with

the rules of courts of equity.”

139

Other than the national-security

exception, all the exceptions outlined in Near relate to speech that falls

outside of the First Amendment’s traditional protective scope.

140

1. Obscenity

Obscenity has no protection under the First Amendment.

141

In Roth

v. United States, the Court held that “obscenity is not within the area

of constitutionally protected speech or press.”

142

The Court reasoned

that “lewd and obscene . . . utterances are no essential part of any

exposition of ideas, and are of such slight social value as a step to truth

that any benefit that may be derived from them is clearly outweighed

by the social interest in order and morality.”

143

Since obscenity provides

such low social value, the Court in Miller v. California

144

affirmed a

broader test to determine what constitutes obscenity. A trier of fact

must determine whether (1) “‘the average person, applying contemp–

orary community standards’ would find that the work, taken as a whole,

appeals to the prurient interest”;

145

(2) the work depicts offensive sexual

conduct; and (3) “the work, taken as a whole, lacks serious literary,

artistic, political, or scientific value.”

146

Given that obscenity receives no constitutional protection, states

have some ability to censor lewd and obscene publications as long as

they take precautions to avoid censoring protected speech. In Times

138. Pentagon Papers, 403 U.S. at 714; Near, 283 U.S. at 716.

139. Near, 283 U.S. at 716.

140. Harper & Row, Publishers, Inc. v. Nation Enters., 471 U.S. 539, 555, 559

(1985); Brandenburg v. Ohio, 395 U.S. 444, 448–49 (1969); Roth v. United

States, 354 U.S. 476, 485 (1957).

141. Roth, 354 U.S. at 485.

142. Id.

143. Id. (quoting Chaplinsky v. New Hampshire, 315 U.S. 568, 571–72 (1942)).

144. 413 U.S. 15 (1973).

145. Id. at 16 (quoting Roth, 354 U.S. at 489).

146. Id.

Case Western Reserve Law Review·Volume 70·Issue 2·2019

Defamatory Political Deepfakes and the First Amendment

441

Film Corp. v. City of Chicago,

147

the Supreme Court upheld a local

ordinance requiring filmmakers to submit their films to a review board

and obtain approval from the city prior to a public exhibition.

148

The

Court clarified this holding in Freedman v. Maryland,

149

where it

affirmed that states could use noncriminal processes, including prior

submission, to censor obscenity in films.

150

However, the prior

submission must “take[] place under procedural safeguards designed to

obviate the dangers of a censorship system.”

151

2. Copyright

A copyright holder has “the right to exclude others from using his

property.”

152

Individuals cannot copyright facts or ideas, but they can

153

Although copyright law creates

a restriction on speech, the Supreme Court held in Harper & Row,

Publishers, Inc. v. Nation Enterprises

154

that the First Amendment does

not protect speech that infringes on another’s copyright.

155

Under the

final injunctions on such terms as it may deem reasonable to prevent

or restrain infringement of a copyright.”

156

Under this statute, modern

courts have been willing to grant not only permanent injunctions but

also preliminary injunctions to protect a plaintiff’s property interests.

157

The practice of granting injunctions in copyright and patent cases

became so pervasive that in 2006 the Supreme Court reminded lower

courts that in granting injunctions, they should not abandon

“traditional equitable considerations.”

158

In eBay Inc. v. MercExchange,

L.L.C., the Supreme Court noted that it has “consistently rejected

invitations to replace traditional equitable considerations with a rule

that an injunction automatically follows a determination that a

147. 365 U.S. 43 (1961).

148. Id. at 46.

149. 380 U.S. 51 (1965).

150. Id. at 58.

151. Id.

152. eBay, Inc. v. MercExchange, L.L.C., 547 U.S. 388, 392 (2006) (quoting

Fox Film Corp. v. Doyal, 286 U.S. 123, 127 (1932)).

153. Lemley & Volokh, supra note 119, at 166.

154. 471 U.S. 539 (1985).

155. Id. at 555–60; Lemley & Volokh, supra note 119, at 150, 166.

156. 17 U.S.C. § 502(a) (2012).

157. Lemley & Volokh, supra note 119, at 150, 158–59.

158. eBay Inc. v. MercExchange, L.L.C., 547 U.S. 388, 392 (2006).

Case Western Reserve Law Review·Volume 70·Issue 2·2019

Defamatory Political Deepfakes and the First Amendment

442

159

Under eBay, a copyright- or patent-

holding plaintiff must satisfy a four-factor test before a permanent

injunction can be granted to protect her intellectual property interest.

160

The four factors are: (1) the plaintiff “has suffered an irreparable

injury”; (2) “remedies available at law . . . are inadequate to compensate

for [the] injury”; (3) “considering the balance of hardships between [the

parties], a remedy in equity is warranted”; and (4) “the public interest

would not be disserved by a permanent injunction.”

161

Although eBay

was a patent case, lower courts have extended its holding to copyright

cases involving preliminary- or permanent-injunction requests.

162

C. Injunctions on Defamatory Speech

Like obscenity and copyrights, defamatory speech has only limited

protection under the First Amendment.

163

Unlike obscenity and

copyrights, however, courts have been hesitant to allow injunctions on

defamatory speech.

164

The Supreme Court did not formally address the

issue of injunctions against non-press related defamation until 2005.

165

In Tory v. Cochran,

166

the Court accepted a case on post-trial

injunctions against defamatory speech, but Cochran’s untimely death

prevented a final resolution of the issue.

167

The Tory case arose out of

a dispute between renowned attorney Johnnie Cochran and Ulysses

Tory, a disgruntled prior client of Cochran’s law firm.

168

Dissatisfied

with Cochran’s services, Tory picketed outside Cochran’s office with

159. Id. at 392–93.

160. Id. at 391.

161. Id.

162. Salinger v. Colting, 607 F.3d 68, 77 (2nd Cir. 2010) (holding that “eBay

applies with equal force (a) to preliminary injunctions (b) that are issued

for alleged copyright infringement”); Peter Lettersese & Assocs. v. World

Inst. of Scientology Enters., 533 F.3d 1287, 1323 (citing eBay’s four-factor

test as the appropriate criteria to evaluate a permanent injunction in a

543 (4th Cir. 2007) (applying eBay’s four-factor test to analyze a

permanent injunction in a copyright case); see also Anthony DiSarro,

Freeze Frame: The Supreme Court’s Reaffirmation of the Substantive

Principles of Preliminary Injunctions, 47 Gonz. L. Rev. 51, 83 (2013).

163. Ashcroft v. Free Speech Coalition, 535 U.S. 234, 245–46 (2002).

164. See Lemley & Volokh, supra note 119, at 149–50.

165. See generally Tory v. Cochran, 544 U.S. 734 (2005).

166. Id.

167. Id. at 737–38; Erwin Chemerinsky, Injunction in Defamation Cases, 57

Syracuse L. Rev. 157, 162 (2007). Chemerinsky served as Tory’s counsel

in Tory. 544 U.S. at 735.

168. Chemerinsky, supra note 167, at 158.

Case Western Reserve Law Review·Volume 70·Issue 2·2019

Defamatory Political Deepfakes and the First Amendment

443

other discontented clients.

169

Cochran sued Tory and the picketers for

defamation. After Tory indicated that he would “continue to engage in

[the defamatory] activity in the absence of a court order,” the California

Superior Court issued an injunction prohibiting Tory from “orally

uttering statements” about Cochran and his law firm in “any public

forum.”

170

Tory challenged the injunction, alleging that it was an

unconstitutional prior restraint on his free-speech rights.

171

The

Supreme Court granted certiorari on the specific question of “[w]hether

a permanent injunction as a remedy in a defamation action, preventing

all future speech about an admitted public figure, violates the First

Amendment.”

172

The case took an unexpected turn when Cochran died one week

after oral argument.

173

Because the injunction against Tory was still in

effect after Cochran’s death, the Court held the case was not moot.

174

However, the Court held that the injunction had lost its underlying

rationale of preventing Tory from coercing Cochran to “pay a ‘tribute’”

for Tory to desist in the defamatory activity.

175

The Court adopted a

very narrow holding: “the injunction, as written, now amounts to an

overly broad prior restraint upon speech, lacking plausible

justification.”

176

Although the Court defined the current injunction as a

prior restraint on speech, it left the door open for injunctions against

defamatory speech in the future: “If . . . injunctive relief may still be

warranted, any appropriate party remains free to ask for such relief.

We express no view on the constitutional validity of any such new relief,

tailored to these changed circumstances, should it be entered.”

177

Although the Supreme Court has not resolved the issue of the

constitutionally of injunctions against defamatory speech, lower courts

have begun to experiment with injunctions in the new age of Internet

defamation.

178

According to a survey by David Ardia of more than 242

decisions involving injunctions against defamatory speech, “at least

fifty-six decisions . . . granted or affirmed [the] injunctions,” with a

169. Id. at 159.

170. Tory, 544 U.S. at 735–36.

171. Chemerinsky, supra note 167, at 161.

172. Tory, 544 U.S. at 735–36; Chemerinsky, supra note 167.

173. Tory, 544 U.S. at 736.

174. Id.

175. Id. at 736–38.

176. Id. at 738.

177. Id. at 738–39.

178. Ardia, supra note 10, at 4.

Case Western Reserve Law Review·Volume 70·Issue 2·2019

Defamatory Political Deepfakes and the First Amendment

444

stark increase in the number of courts granting injunctions after 2000.

179

Both federal

180

and state

181

courts have permitted injunctions against

defamatory speech, although courts vary widely in defining the

appropriate scope of those injunctions.

182

Courts tend to rely on three

primary justifications when granting permanent injunctions against

defamatory speech: “(1) the speech impugned the plaintiff’s property

interest; (2) the defendant engaged in a continuing course of conduct

that caused the plaintiff harm; or (3) the speech had been adjudged to

be defamatory.”

183

The most common reason cited for upholding

injunctions against speech is that such speech had already been

adjudged to be defamatory, therefore an injunction against it would not

be a prior restraint on protected speech.

184

Federal courts are split on the issue of injunctions against defam–

atory speech. Although some circuits have taken the traditional view

that “equity does not enjoin a libel or slander,”

185

the Fifth, Sixth, and

Ninth Circuits have permitted at least limited injunctions against

defamatory or fraudulent speech.

186

Both the Fifth and Sixth Circuits

permitted injunctions against defamatory speech only if the injunctions

179. Id. at 42.

180. See infra text accompanying notes 186–190 (highlighting that, although

federal courts are split, some courts such as the Sixth Circuit issue limited

injunctions for defamatory speech).

181. Ohio, Georgia, Minnesota, California, Kentucky, and Montana have

adopted rules allowing injunctions on speech that courts deem

defamatory. See Ardia, supra note 10, at 48–50; O’Brien v. Univ. Cmty.

Tenants Union, 327 N.E.2d 753, 755 (Ohio 1975); Retail Credit Co. v.

Russell, 218 S.E.2d 54, 62 (Ga. 1975); Advanced Training Sys., Inc. v.

Caswell Equip. Co., 352 N.W.2d 1, 11 (Minn. 1984); Balboa Island Vill.

Inn v. Lemen, 156 P.3d 339, 348–49 (Cal. 2007); Hill v. Petrotech Res.

Corp., 325 S.W.3d 302, 309 (Ky. 2010); St. James Healthcare v. Cole, 178

P.3d 696, 703 (Mont. 2008) (adding that not every injunction that

amounts to a prior restraint is “impermissible”).

182. Ardia, supra note 10, at 52–57 (outlining four types of injunctions utilized

by various federal and state courts).

183. Id. at 43–44.

184. Id. at 48.

185. Cmty. for Creative Non-Violence v. Pierce, 814 F.2d 663, 672 (D.C. Cir.

1987) (quoting Kukatush Mining Corp. v. SEC, 108 F.Supp. 508, 510–11

(D.D.C. 1961)); Metro. Opera Ass’n v. Local 100, Hotel Emps. & Rest.

Emps. Int’l Union, 239 F.3d 172, 177 (2d Cir. 2001).

186. See San Antonio Cmty. Hosp. v. S. Cal. Dist. Council of Carpenters, 125

F.3d 1230, 1239 (9th Cir. 1997) (upholding an injunction against

fraudulent and defamatory signs held by union members during a labor

dispute); Brown v. Petrolite Corp., 965 F.2d 38, 51 (5th Cir. 1992);

Lothschuetz v. Carpenter, 898 F.2d 1200, 1208–09 (6th Cir. 1990)

(Wellford, J., concurring in part and dissenting in part).

Case Western Reserve Law Review·Volume 70·Issue 2·2019

Defamatory Political Deepfakes and the First Amendment

445

were limited to speech that had been previously adjudged to be

defamatory.

187

In Lothschuetz v. Carpenter, Carpenter, a company

owner who had previously been involved in administrative and judicial

proceedings with two attorneys, wrote and sent defamatory letters

about the attorneys to the President of the attorneys’ workplace.

188

The

attorneys sued Carpenter for defamation and sought a permanent

injunction preventing Carpenter from publishing further defamatory

statements.

189

Through dual concurring opinions, the Sixth Circuit

granted a limited injunction emphasizing that because Carpenter’s

defamatory statements were “frequent and continu[ous],” an injunction

was an appropriate remedy to “prevent future injury to [the attorneys’]

personal reputation[s].”

190

Other circuit courts have not directly addressed the issue of

whether any injunction against defamatory speech constitutes an

unconstitutional prior restraint; but they have rejected proposed

injunctions on the basis that they were overbroad.

191

The First Circuit

recently took this approach in Sindi v. El-Moslimany.

192

There, the First

Circuit struck down an injunction prohibiting an individual from

“publishing ‘orally, in writing, through direct electronic commun–

ications, or by directing others to website or blogs reprinting’ six

statements that the district court concluded were defamatory.”

193

The

trial court granted a permanent injunction against a woman engaged in

an online campaign to smear and defame a scientist whom she believed

was engaged in an affair with her husband.

194

The circuit court

effectively applied a strict-scrutiny test to the injunction.

195

Since the

trial court’s injunction was not “as narrowly tailored as possible to

187. Petrolite Corp., 965 F.2d at 51 (ordering the district court to “narrow its

previous language so as to enjoin the dissemination of information relating

to . . . the subject of the underlying suit”); Lothschuetz, 898 F.2d at 1208–

09 (Wellford, J., concurring in part and dissenting in part) (limiting the

injunction “to the statements which have been found in this and in prior

proceedings to be false and libelous”).

188. Lothschuetz, 898 F.2d at 1203.

189. Id. at 1203–04.

190. Id. at 1208–09 (Wellford, J., concurring in part and dissenting in part).

191. See Sindi v. El-Moslimany, 896 F.3d 1, 34 (1st Cir. 2018); McCarthy v.

Fuller, 810 F.3d 456, 461–62 (7th Cir. 2015).

192. Sindi, 896 F.3d 1, at 27–36.

193. Id. at 12, 34.

194. Id. at 11–12.

195. Id. at 30, 37 n.16.

Case Western Reserve Law Review·Volume 70·Issue 2·2019

Defamatory Political Deepfakes and the First Amendment

446

avoid censoring protected speech,” the circuit court held that it

amounted to an unconstitutional prior restraint.

196

In dissent, Judge Barron argued against the “application of strict

scrutiny.”

197

Barron relied on Schenck v. Pro-Choice Network of

Western New York

198

and Madsen v. Women’s Health Center

199

assert that the Supreme Court has upheld injunctions against speech as

a prophylactic measure when the enjoined parties were likely to

continue to engage in the unprotected conduct.

200

Barron argued that

the appropriate standard of review for injunctions against defamatory

speech should be one that “burden[s] no more speech than necessary to

serve a significant government interest,”

201

not that the injunction is

“the least restrictive means of achieving a compelling state interest.”

202

Thus, courts have left open two key questions: (1) whether all

injunctions against defamatory speech constitute unconstitutional prior

restraints; and (2) if some injunctions are permissible, what is the

appropriate scope of injunctions against defamatory speech so that the

injunctions comport with the First Amendment. In an effort to explore

whether some injunctions may be permissible under the First

Amendment, David Ardia sorted injunctions against defamatory speech

into four types based on their restrictiveness.

203

Type I injunctions

“prohibit[] a party from making any statements about the plaintiff.”

204

These injunctions are extremely broad and highly unlikely to pass a

First Amendment challenge. Type II injunctions “prohibit a party from

publishing any defamatory statements about the plaintiff.”

205

Type II

injunctions are also likely to be overbroad because “they are not precise

enough to put the defendant on notice as to what speech will violate

the injunction.”

206

Although both Type I and Type II injunctions target

individuals who have engaged in defamatory conduct in the past, they

overreach and effectively deprive the defendants of the “breathing

196. Id. at 35.

197. Id. at 45.

198. 519 U.S. 357, 381–82 (1997).

199. 512 U.S. 753, 768–71 (1994).

200. Sindi, 896 F.3d at 45–46.

201. Id. at 45 (quoting Schenck, 519 U.S. at 372).

202. Id. (quoting McCullen v. Coakley, 573 U.S. 464, 478).

203. Ardia, supra note 10, at 52.

204. Id.

205. Id. at 53.

206. Id.

Case Western Reserve Law Review·Volume 70·Issue 2·2019

Defamatory Political Deepfakes and the First Amendment

447

space” necessary to engage in public debate.

207

Such restrictions amount

to unconstitutional prior restraints on speech.

Type III injunctions “prohibit a party from publishing certain

enumerated statements about the plaintiff without limiting the

injunction to the specific statements that have been found to be

defamatory.”

208

Type III injunctions are probably the closest to the

injunction granted by the trial court in Sindi.

209

Type III injunctions

are attractive to plaintiffs who have been defamed by spoken words or

written commentaries because the injunctions are flexible enough to

capture potential defamatory speech that has not yet been published.

But some courts may consider Type III injunctions unconstitutional

prior restraints because they cover speech that has not yet been

adjudged defamatory. In addition, Type III injunctions are not be the

least-restrictive remedy in most cases, so they are not likely to survive

a strict-scrutiny analysis. However, some courts have been willing to

permit such injunctions if the defendant’s defamatory conduct is either

ongoing or coercive.

210

In cases where the defendant admits that she

plans to continue engaging in defamatory conduct, a court may find

that preventing harassment serves as a sufficiently important interest

to permit an injunction.

211

If, however, a specific defamatory work could

be targeted with precision, Type III injunctions may be overly broad.

Thus, we are left with Type IV injunctions, which “only prohibit[]

further publication, or orders the removal of the specific statements a

court or jury has found are defamatory.”

212

Type IV injunctions are the

narrowest possible injunctive remedy; thus, if a plaintiff can articulate

a compelling interest for the injunction, it may even clear strict

scrutiny’s high hurdle. But Type IV injunctions are exceedingly difficult

to apply against spoken words or specific passages of written

commentary because defamatory comments made in this manner are

often very contextual.

213

Type IV injunctions are more easily applied in

cases where a video or a complete work is posted on the Internet. The

207. N. Y. Times Co. v. Sullivan, 376 U.S. 254, 271–72 (1964).

208. Ardia, supra note 10, at 54.

209. Sindi v. El-Moslimany, 896 F.3d 1, 12 (1st Cir. 2018) (enjoining the

publication, through oral and written means, electronic communications,

or direction to publications, of six statements it held to be defamatory).

210. Id. at 44–45 (Barron, J., dissenting); Lothschuetz v. Carpenter, 898 F.2d

1200, 1208–09 (6th Cir. 1990) (Wellford, J., concurring in part and

dissenting in part).

211. Sindi, 896 F.3d at 44 (Barron, J., dissenting); Lothschuetz, 898 F.2d at

1208–09 (Wellford, J., concurring in part and dissenting in part).

212. Ardia, supra note 10, at 56.

213. Sindi, 896 F.3d at 30 (noting the injunction cannot stand “principally

because of its failure to account for contextual variation”).

Case Western Reserve Law Review·Volume 70·Issue 2·2019

Defamatory Political Deepfakes and the First Amendment

448

injunction can be tailored to that specific work in an almost surgical

fashion, removing only the files or words deemed unprotected while

leaving intact the remaining protected speech. In many ways, these very

limited injunctions against defamatory speech can work in the same

way as injunctions against obscenity by targeting a specific expressive

work. In dicta, the First Circuit mused on this distinction, noting that

the “obscenity doctrine proscribes specific expressive works (such as

books or movies) that . . . lack ‘serious literary, artistic, political, or

scientific value,’” while “injunction[s] that prevent[] in perpetuity the

utterance of particular words and phrases after a defamation trial is

quite a different matter.”

214

Thus, courts have not developed a consensus on how to address

injunctions against defamatory speech. Even so, examples of permissible

injunctions in obscenity and copyright cases, as well as Ardia’s

framework for injunctions against defamatory speech, can serve as a

basis for determining whether injunctions against defamatory political

deepfakes might survive a First Amendment challenge.

VI. Injunctions on Defamatory Deepfakes

Deepfakes are strikingly different from many other forms of

defamatory speech in both the manner they are created and in their

form of expression. First, unlike other defamatory speech, all of a

deepfake’s content is inherently false, so there is a higher presumption

of a culpable mental state on the part of the deepfake creator. Second,

deepfakes have little social value because they are inherently false and

they undermine the public’s ability to distinguish what is true from

what is false. Finally, it is easier to enjoin the expressive form of

deepfakes, as opposed to enjoining their underlying ideas, because

deepfake creators need the assistance of artificial intelligence to

generate deepfakes.

In most defamatory speech cases involving spoken or written words,

it is highly unlikely that all of the expression is untrue. This is exactly

the case, however, with deepfakes: every frame is a falsehood. As

defined above, deepfakes “alter the content of an original video, image,

and/or audio file by face swapping or scenario alterations.”

215

Unlike

with spoken or written words, when a court enjoins a deepfake, it

eliminates the possibility that the content could be republished in the

same form (video, photo, or audio) in a true manner. Even if the

publisher changes a few frames in the deepfake, the remainder would

still be false. At best, reproducing a deepfake without some offending

frames still creates false, but non-defamatory, content. In United States

214. Id. at 33.

215. See supra Part I.C.

Case Western Reserve Law Review·Volume 70·Issue 2·2019

Defamatory Political Deepfakes and the First Amendment

449

v. Alvarez,

216

the Supreme Court upheld some First Amendment

protections even for false speech.

217

Still, the value of protecting false

speech is not sufficient to outweigh important or compelling

governmental interests, such as free and fair elections or national

security.

In addition, a deepfake creator intentionally alters every frame of a

video to intentionally distort reality. Such a calculated act creates a

much stronger presumption that a deepfake creator has a culpable

mental state as compared to a publisher of an angry social-media post

who slips in an untrue detail about another person. Blog-post creators

may credibly deny that they knew a specific detail of their blog was

false when it was published; but deepfake creators will have a much

more difficult time convincing a fact-finder that they were unaware they

were creating and publishing a deepfake. Thus, the way deepfakes are

created brings deepfake cases in line with copyright or obscenity cases

in that there is more evidence that the infringing material was created

by someone with a guilty mental state. In copyright cases, it is

exceedingly difficult for a copier to convince a fact-finder that she

coincidentally replicated another author’s work word for word.

Similarly, in obscenity cases, it is unlikely that a jury would find that

a pornographic video’s publisher was unaware that a group-sex video

she published would be considered porn. Indeed, given the current

technological skills needed for deepfake creation, it might be harder for

plaintiffs to prove a culpable mental state in many copyright and

obscenity cases than it would be in a deepfake case.

Second, defamatory deepfakes, similar to obscenity, have little

social value. There are some positive uses for non-defamatory deepfakes,

such as artistic expression or satire,

218

but the underlying content of a

defamatory deepfake will always be (at minimum) a recklessly created

falsehood. Combining the deepfake’s ability to spread falsehoods with

the video medium’s generally perceived truthfulness undermines the

public’s ability to believe its own eyes. Although our society can tolerate

some protections for falsehoods, once a deepfake crosses the line into

defamation it is stripped of all social value. Defamatory deepfakes do

not contribute to public discourse; instead they distort open-forum

principles by undermining the concept of truth itself. Similar to

216. 567 U.S. 709 (2012).

217. See id. at 719–720 (overturning the Stolen Valor Act, which made it a

crime to falsely claim receipt of a military decoration, on the ground that

“falsity alone” is not enough to bring speech outside the First

Amendment).

218. See supra Part IV (explaining how satire or parody can be a good-faith

defense against deepfake accusations).

Case Western Reserve Law Review·Volume 70·Issue 2·2019

Defamatory Political Deepfakes and the First Amendment

450

obscenity, any benefit derived from defamatory deepfakes “is clearly

outweighed by the social interest in order and morality.”

219

Third, and similar to copyright, it is easier to enjoin a deepfake as

a “form of expression” than it would be for most other forms of

defamatory speech.

220

Deepfakes, by definition, must be created “with

the use of deep learning.”

221

Deepfakes create a specific type of

electronic file (either video, photo, or audio), and a court has the ability

to make the distinction between enjoining the specific form of

expression (i.e., content created with the aid of artificial intelligence)

and enjoining the underlying ideas or speech in the video. Although it

might seem that enjoining only the form of expression does not go far

enough to protect political figures from defamatory speech, such an

argument is unavailing. The novel threat of deepfakes is that the

manner in which they are created—and the manner in which they are

viewed—makes them more likely to be perceived as real. For “seeing is

believing,” and realistic-looking videos are generally more likely to be

believed than written commentaries.

222

Thus, enjoining the form of

expression, as opposed to the underlying speech, at worst returns

political figures to the status quo.

VII. Appropriate Scope for Injunctions Against

Political Defamatory Deepfakes

Although defamation may be the appropriate legal theory by which

political figures should address deepfakes in a court of law, obscenity

and copyright law provide the appropriate remedies. Similar to the

obscenity test defined in Miller, the standard to determine whether a

deepfake is defamatory should be: if “the average person, applying

contemporary community standards would find the work, taken as a

whole” to be defamatory and “lack[ing] serious literary, artistic,

political, or scientific value,”

223

then the deepfake is defamatory. If the

entire work is not considered as a whole, a deepfake creator could

change just one frame in the offending video and repost it to the

Internet in its entirety. Once the “work as a whole” has been adjudged

219. Roth v. United States, 354 U.S. 476, 485 (1957) (quoting Chaplinsky v.

New Hampshire, 315 U.S. 568, 571–72 (1942)). But see Gertz v. Robert

Welch Inc., 418 U.S. 323, 341 (1974) (explaining that the “First

Amendment requires that we protect some falsehood in order to protect

speech that matters”).

220. See Lemley & Volokh, supra note 119, at 167.

221. See supra Part I.C.

222. Brundage et al., supra note 56, at 46.

223. Miller v. California, 413 U.S. 15, 24 (1973).

Case Western Reserve Law Review·Volume 70·Issue 2·2019

Defamatory Political Deepfakes and the First Amendment

451

defamatory, then the court can craft a narrow injunction related to the

specific deepfake.

Erwin Chemerinsky, who served as counsel for Ulysses Tory at the

Supreme Court, argued against injunctions of defamatory speech: “Any

effective injunction will be overbroad and any limited injunction will be

ineffective.”

224

Given the properties of a deepfake, it is possible to craft

an injunction that is both narrowly tailored and effective. For instance,

in copyright cases, courts have the original work to compare against

the potentially infringing content. Courts can specifically identify

whether a particular expression sufficiently mimics that of the

copyrighted work.

225

Similarly, with deepfakes, courts can use certain

technology to see whether republications of a defamatory deepfake are

the same as the original video that it already deemed defamatory.

226

Even if courts choose to follow the current legal framework of

injunctions under defamation law, very narrowly tailored injunctions—

such as the Type IV injunctions outline in Part V.C.—would likely be

effective at removing offending material without disturbing protected

speech. A court could craft an injunction such as the one below:

Defendants are permanently enjoined from publishing the

deepfake, (describe deepfake file name here), adjudged

defamatory by this court, either in its entirety or in part on any

public medium (including, but not limited to, the Internet,

television, and/or radio). Publication in part shall include the

publication of any frame, series of frames, and/or audio snippet(s)

from (describe deepfake file name here).

Such an injunction is broad enough to prevent further harm from

the specific deepfake, while still being narrow enough to avoid censoring

constitutionally protected speech. Given deepfakes’ special

characteristics and risks, they serve as a unique example that some

injunctions on defamatory speech may comport with the First

Amendment.

VIII. Injunctions and Third-Party Providers

Even if an injunction against a deepfake is found to be

constitutional, under the current statutory scheme only the creator can

be held liable for not removing the offending content from the

224. Chemerinsky, supra note 167, at 171.

225. Lemley & Volokh, supra note 119, at 159–61.

226. Will Knight, The Defense Department has Produced the First Tools

for Catching Deepfakes, Tech. Rev. (Aug. 7, 2018), https://www.

technologyreview.com/s/611726/the-defense-department-has-produced-the-

first-tools-for-catching-deepfakes/ [https://perma.cc/P6GQ-EEDU].

Case Western Reserve Law Review·Volume 70·Issue 2·2019

Defamatory Political Deepfakes and the First Amendment

452

Internet.

227

Under section 230 of the Communications Decency Act of

1996 (“CDA”),

228

Internet service providers (“ISPs”) are not subject to

civil liability for the content posted to their webpages by a third

party.

229

Congress adopted the CDA wake of the Stratton Oakmont v.

Prodigy Services,

230

where the Supreme Court of New York held that

an interactive ISP that monitored its services should be considered a

publisher of defamatory content, even though a third party posted the

content on the ISP’s webpage.

231

Congress feared that exposing ISPs to

civil liability would discourage self-censorship. Thus, Congress provided

immunity from civil liability to ISPs for content posted by third parties,

even if the ISP monitors or modifies the content.

232

Courts have

interpreted the CDA broadly, insulating “interactive computer

services” from all civil liability (except copyright infringement) for

content posted on their websites.

233

The blanket immunity provided to ISPs by CDA § 230 could pose

a major issue for defamatory political deepfakes because many deepfake

creators will likely be unreachable by the judicial system. Crafty

deepfake creators may post their creations under fake usernames and

leave an insufficient digital trail to be identified and brought to court.

Also, a multitude of deepfake creators may reside outside the physical

jurisdiction of the United States.

234

Many legal commentators have

addressed the issues of ISP immunity under the CDA § 230.

235

227. 47 U.S.C. § 230(c)(1) (2012).

228. Id. § 230.

229. Id.

230. No. 31063/94, 1995 WL 323710 (N.Y. Sup. Ct. May 24, 1985).

231. Id. at *4–5.

232. Kate Klonick, The New Governors: The People, Rules, and Processes

Governing Online Speech, 131 Harv. L. Rev. 1598, 1602 (2018).

233. Jonathan Band & Matthew Schruers, Safe Harbors Against the Liability

Hurricane: The Communications Decency Act and The Digital

Millennium Copyright Act, 20 Cardozo Arts & Ent L.J. 295, 295

(2002).

234. Chesney & Citron, supra note 3, at 44. Some individuals may have even

been directed by a foreign state to meddle in American politics. Id.

235. See generally Danielle Citron, Hate Crimes in Cyberspace (2014)

(discussing the effects of cyber harassment and necessary legal reforms,

including addressing CDA-created immunity); Danielle Keats Citron &

Benjamin Wittes, The Internet Will Not Break: Denying Bad Samaritans

Section 230 Immunity, 86 Fordham L. Rev. 401 (2017); Joseph G.

Marano, Note, Caught in the Web: Enjoining Defamatory Speech that

Appears on the Internet, 69 Hastings L.J. 1311 (2018) (arguing that

“the overbroad immunity courts have read into section 230 of the CDA,

necessitate[s] a remedy”).

Case Western Reserve Law Review·Volume 70·Issue 2·2019

Defamatory Political Deepfakes and the First Amendment

453

Deepfakes are merely another addition to a growing list of concerns

created by blanket immunity to ISPs.

236

Because deepfakes create unique issues and threats, Congress

should create an exception for them, similar to exception created for

(“DMCA”).

237

The DCMA provides ISPs with a “safe harbor” to limit

ISPs’ liability for monetary damages for copyrighted materials hosted

on their websites.

238

ISPs can, however, be held in contempt for failing

to comply with an injunction if they do not comply with several general

requirements.

239

ISPs “must adopt, implement, and inform users of a

policy providing for the termination of repeat infringers.”

240

In addition,

ISPs must employ “standard technical measures” to detect copyrighted

works.

241

Most importantly for copyright holders, to be eligible for the

“safe-harbor” protections, ISPs must, “upon notification of claimed

infringement . . . respond[] expeditiously to remove, or disable access

to, the material that is claimed to be infringing or to be the subject of

infringing activity.”

242

This provision is often referred to as “notice and

takedown” procedure.

243

Extending the protections of the DMCA to cover defamatory

deepfakes would significantly reduce their potential harm. Even if a

political figure was unable to find the deepfake’s original creator, she

could bring a defamation case against the unnamed creator. If political

figure could prove that the video is a deepfake, and the court finds the

video to be defamatory, the political figure could obtain a default

judgment, including a permanent injunction (like the one described in

Part VII). The political figure could then send a takedown notice to the

ISP, backed by the full weight of judicial authority.

Although such a process allows the defamatory video to be posted

longer than any political figure would like, the process provides a

multitude of other protections. First, many ISPs actively self-police,

and upon notice that a political figure has filed a high-profile lawsuit

alleging a video to be a deepfake, they would likely take down the video

prior to any court proceedings either out of a sense of corporate

236. Chesney & Citron, supra note 3, at 38–39.

237. 17 U.S.C. § 512 (2012); Band & Schruers, supra note 233.

238. 17 U.S.C. § 512(i) (2012); Band & Schruers, supra note 233, at 304.

239. 17 U.S.C. § 512(j) (2012); Band & Schruers, supra note 233, at 304.

240. 17 U.S.C. § 512(i) (2012); Band & Schruers, supra note 233, at 304.

241. 17 U.S.C. § 512(i) (2012); Band & Schruers, supra note 233, at 304.

242. 17 U.S.C. § 512(c)(1)(C) (2012); Band & Schruers, supra note 233, at

305.

243. Jennifer M. Urban, Takedown in Two Worlds: An Empirical Analysis, 64

Case Western Reserve Law Review·Volume 70·Issue 2·2019

Defamatory Political Deepfakes and the First Amendment

454

responsibility or for economic reasons.

244

Even if the ISP does not self-

police, and the video is allowed to remain on the Internet throughout

the duration of the suit, simply giving political figures the ability to sue

allows them to signal that they are contesting the validity of a posted

video. The political figures can then publicize their lawsuit, and an

attentive press would likely report on that lawsuit. The political figure’s

lawsuit has the secondary effect of warning the press to proceed with

caution on their reporting of the video’s contents.

Certainly, the DMCA process is not perfect,

245

but it is the only

time-tested process for dealing with individually policing and removing

non-protected speech posted on the Internet. Due to ISPs’ familiarity

with the DMCA, extending it to deepfakes should create only minimal

additional costs for ISPs. Although deepfakes will become progressively

easier to create, they will likely never be as pervasive as copyright

infringement. Also because, at least for the time being, creating a

deepfake leaves a digital trace, deepfakes should be easier for ISPs to

detect than copyright infringements.

246

Given the minimal additional

cost on ISPs, extending the DMCA to defamatory deepfakes would be

a cost effective way to deal with the serious threats they pose to society.

Conclusion

It is clear that defamatory political deepfakes could cause serious

harms to individuals and society. Such harms, however, must be

balanced against protecting freedom of speech under the First

Amendment. Courts have rightfully been skeptical of restraints on

expression, especially if they have the possibility of restricting protected

speech. But in the case of deepfakes, courts have an opportunity to

specifically define the form of expression that is defamatory. Courts also

bear a far lower risk of misclassifying speech as defamatory because

deepfakes are purposefully designed to be false. Thus, courts should

draw lessons from both obscenity and copyright law to allow some

narrowly crafted permanent injunctions against deepfakes. Such

narrowly crafted injunctions will help prevent the spread of

misinformation that might undermine elections or trust in public

officials while providing the least restrictive means possible of limiting

244. See Klonick, supra note 232, at 1625–30.

245. See generally Jacqueline Lipton, The Law of Unintended Consequences:

The Digital Millennium Copyright Act and Interoperability, 62 Wash. &

Lee L. Rev. 487 (2005); Methaya Sirichit, Catching the Conscience: An

Analysis of the Knowledge Theory Under § 512(C)’s Safe Harbor & the

Role of Willful Blindness in the Finding of Red Flags, 23 Alb. L.J. Sci.

Constitutionality: How the Digital Millennium Copyright Act Goes Too

Far, 79 Notre Dame L. Rev. 2115 (2004); Urban, supra note 243, at

486.

246. Knight, supra note 226.

Case Western Reserve Law Review·Volume 70·Issue 2·2019

Defamatory Political Deepfakes and the First Amendment

455

expression. In addition, Congress should consider expanding the DMCA

or other similar legislation to include defamatory deepfakes. Only

through such action will political figures (and society, in general) have

a legitimate judicial solution to the harms created by defamatory

political deepfakes.

Returning to this Note’s original hypothetical, if Trump had an

opportunity to counter the defamatory deepfake through a judicial

process that provided him with an equitable remedy to remove the

offending content, then some of the video’s potential harms would be

at least curbed, if not eliminated. Instead of simply denying the video’s

truth, using the judicial process to counter and remove such content

would bring more legitimacy to deepfake target’s denial, and

consequently, to the political system in general. Such a hypothetical

could happen to any political figure at any governmental level. As a

society, regardless of our individual political beliefs, we need a more

legitimate system for combating this kind of misinformation, especially

when it comes in such a virulent form as a deepfake video. The judicial

system, drawing on lessons learned regarding other types of limited

protected speech, could provide this solution.

Jessica Ice

†

† J.D. Candidate, 2020, Case Western Reserve University School of Law;

winner of the 2019 Sindell Torts Prize. The author would like to thank

the editors of Case Western Reserve Law Review, Volume 70, for their

careful edits and Professor Emeritus Jonathan L. Entin for his helpful

insight on First Amendment jurisprudence. Additional thanks to my

husband, Stephen Mucher, for providing the inspiration for this topic.