The Landscape of Deepfake AI Legislation
**Attorney Advertising**

BACKGROUND
In 2012, advances in deep learning and convolutional neural networks (CNNs) allowed machines to better categorize images.¹ The public gained access to these technological innovations in 2017.² These techniques experienced a significant improvement in 2022 with Diffusion models and the release of applications, such as Stable Diffusion and Midjourney, that provided people with the capacity to rapidly generate realistic images.³ Beyond just image generation, the increasing quality of human voice re-enactment has endowed upon people the ability to create synthetic voices that can emulate real people.⁴ This technology has already been utilized for political purposes with a notable 2020 instance of an emulation of President Barack Obama’s voice and a 2021 instance of Indian politician Manoj Tiwar posting a deepfake video of himself speaking in a Hindi dialect.⁵

Due to the novelty of this deepfake technology, it remains a relatively under-legislated area. This memo will provide an overview of the different federal and state-level efforts that are being undertaken to address this nascent technology and the problems associated with it.

Non-Consensual Explicit Image Sharing
In January 2024, a bipartisan group of Senators led by Sen. Richard Durbin (D – IL) introduced the Disrupt Explicit Forged Images and Non-Consensual Edits Act (DEFIANCE Act), which is aimed at addressing AI generation of non-consensual images.⁶ The bill’s proposal came in the aftermath of inappropriate AI images of Taylor Swift being posted on social media.⁷ If passed, the DEFIANCE Act would allow victims to seek relief by means such as injunctive relief, punitive damages, and any other cognizable damages present; specifically, the DEFIANCE Act would function by amending the Violence Against Women Act (VAWA) to create a civil private right of action.⁸ As of June 2024, the DEFIANCE Act has not advanced past the Senate Committee on the Judiciary.⁹

At the state level, Minnesota and New York have both passed laws to criminalize deepfake explicit images and to allow victims to sue creators of AI images using their likenesses.¹⁰ While Georgia, Hawaii, Indiana, Texas, and Virginia have all passed laws that criminalize nonconsensual deepfake explicit images, they do not have any additional laws that grant victims the right to sue.¹¹ Meanwhile, California and Illinois have passed laws that grant rights to victims, but they have not passed any bans on the creation of these images.¹² Florida, South Dakota, and Washington have passed laws that criminalize the use of deepfakes in explicit images involving minors, but have not addressed these images in cases involving adults.¹³ Missouri, New Jersey, and Ohio all have proposed legislation that addresses the issue of AI deepfake technology in creating explicit images, but no such law has been fully enacted in these states.¹⁴

Deepfake Labelling
In March 2024, the Protecting Consumers from Deceptive AI Act was introduced in the House of Representatives and, if passed, would create standards that determine how AI-generated content should be labeled to users.¹⁵ The Protecting Consumers from Deceptive AI Act contains four key facets: (1) directing the National Institute of Standards and Technology (NIST) to develop standards that can identify and label AI-generated content via technical means, (2) mandating that generative AI developers include machine-readable disclosures in their generated content while allowing users to add metadata with their content, (3) ensuring that online platforms post disclosures with any posted AI-generated content, and (4) working with leading AI companies, experts, and global stakeholders to foster commitments to utilize AI for ethical purposes.¹⁶ Any violation of this Act would be treated as a Federal Trade Commission (FTC) violation concerning unfair or deceptive acts or practices.¹⁷ As of June 2024, the Protecting Consumers from Deceptive AI Act was referred to the House Subcommittees on Innovation, Data, and Commerce, and has not progressed beyond that point.¹⁸

On September 20, 2023, the DEEPFAKES Accountability Act was introduced, and it aims to protect national security while providing recourse for victims of harmful deepfake technology.¹⁹ The bill would require any deepfake content distributed online to include clear disclosures indicating the use of altered audio or visual elements.²⁰ It would impose both criminal and civil penalties for non-compliance, including fines and imprisonment.²¹ Additionally, the Act would establish a private right of action for individuals depicted in harmful deepfakes, allowing them to seek damages and injunctive relief.²² Enforcement would be handled by the Attorney General and the FTC, and would include measures for privacy protection and cooperation with private sector platforms to detect and manage deepfakes.²³ As of June 2024, the bill was referred to the House Subcommittee on Emergence Management and Technology and has not advanced beyond that point.²⁴

On May 1, 2024, the Utah Artificial Intelligence Policy Act went into effect and established several requirements for labeling and disclosure when interacting with AI.²⁵ Specifically, it mandates that any person using generative AI in connection with regulated occupations or consumer transactions must clearly disclose the use of AI.²⁶ This includes verbal disclosure at the start of oral interactions and electronic disclosure before written exchanges.²⁷ Additionally, the Act set up the Office of Artificial Intelligence Policy to oversee these regulations, enforce disclosures, and administer an AI learning laboratory to evaluate AI technologies and their regulatory impacts.²⁸

The upcoming 2024 elections have inspired legislation at the state-level that requires labeled disclaimers for AI-generated advertisements. New Mexico recently enacted HB 182, which mandates disclaimers for “materially deceptive” AI-generated ads.²⁹ Similarly, Florida signed HB 919 into law, which requires disclaimers for AI-generated political advertisements.³⁰ Wisconsin passed a law that would fine political candidates and groups $1,000 for not properly disclosing the use of AI in advertisements.³¹ While California has not yet passed AI disclosure laws, there are at least 4 proposed measures in the California State Legislature that pertain to AI watermarking and disclosure.³² In New York, a law was passed that holds radio stations liable if they knowingly air a political ad that contains AI-generated materials without disclosing it to the public.³³ For this New York law, radio stations must announce at the beginning and end of an advertisement that the ad contains manipulated media.³⁴

CONCLUSION
As deepfake technology continues to advance and permeate various aspects of society, legislative responses at both federal and state levels are crucial to address the associated risks and challenges. While the DEFIANCE Act, the Protecting Consumers from Deceptive AI Act, and the DEEPFAKES Accountability Act represent significant federal efforts to regulate deepfake content, they have not yet progressed beyond committee referrals. In contrast, several states have enacted laws addressing these issues. Minnesota and New York have criminalized deepfake explicit images and provided avenues for victims to sue creators, while California and Illinois offer legal recourse without banning the creation of such images. States like Florida, South Dakota, and Washington have targeted deepfakes involving minors. The Utah Artificial Intelligence Policy Act and recent state-level legislation on AI-generated political advertisements, including New Mexico’s HB 182 and Florida’s HB 919, highlight successful state initiatives. However, inconsistencies in legal protections persist across states.

¹ Johannes Langguth et al., Don’t Trust Your Eyes: Image Manipulation in the Age of DeepFakes, 6 FRONTIERS COMMC’N 1, 2 (2021).
² TSee Alex Krizhevsky et al., ImageNet Classification with Deep Convolutional Neural Networks, 60 Commc’ns ACM 84 (2017).
³ See LANGGUTH, supra note 1.
⁴ Rongjie Huang et al., AudioGPT: Understanding and Generating Speech, Music, Sound, and Talking Head, 38 Proc. AAAI Conf. on A.I. 23802 (2024).
⁵ Rongjie Huang et al., AudioGPT: Understanding and Generating Speech, Music, Sound, and Talking Head, 38 Proc. AAAI Conf. on A.I. 23802
⁶ Solcyré Burga, Deepfakes Are at the Center of A New Federal Bill, TIME (2024), https://time.com/6590711/deepfake-protection-federal-bill/ (last visited Jun 12, 2024).
⁷ Id.
⁸ Disrupt Explicit Forged Images and Non-Consensual Edits (DEFIANCE) Act of 2024, Democratic Women’s Caucus (2024), http://democraticwomenscaucus.house.gov/legislation/disrupt-explicit-forged-images-and-non-consensual-edits-defiance-act-2024 (last visited Jun 12, 2024).
⁹ Sen. Richard J. Durbin, S.3696 – 118th Congress (2023-2024): DEFIANCE Act of 2024, (2024), https://www.congress.gov/bill/118th-congress/senate-bill/3696/all-info (last visited Jun 12, 2024).
¹⁰ Geoff Mulvihill, What to Know about How Lawmakers Are Addressing Deepfakes like the Ones That Victimized Taylor Swift, AP News (2024), https://apnews.com/article/deepfake-images-taylor-swift-state-legislation-bffbc274dd178ab054426ee7d691df7e (last visited Jun 12, 2024).
¹¹ Id.; Madyson Fitzgerald, States Race to Restrict Deepfake Porn as It Becomes Easier to Create • Stateline, Stateline (Apr. 10, 2024), https://stateline.org/2024/04/10/states-race-to-restrict-deepfake-porn-as-it-becomes-easier-to-create/ (last visited Jun 12, 2024).
¹² See Mulvihill, supra note 10.
¹³ See Fitzgerald, supra note 11.
¹⁴ Id.
¹⁵ Rep. Anna Eshoo, Rep. Eshoo Introduces Bipartisan Bill to Label Deepfakes, Congresswoman Anna G. Esho, California’s 16th Congressional District (2024), http://eshoo.house.gov/media/press-releases/rep-eshoo-introduces-bipartisan-bill-label-deepfakes (last visited Jun 12, 2024).
¹⁶ Id.
¹⁷ Id.
¹⁸ Rep. Anna Eshoo, H.R.7766 – 118th Congress (2023-2024): Protecting Consumers from Deceptive AI Act, (2024), https://www.congress.gov/bill/118th-congress/house-bill/7766/text (last visited Jun 12, 2024).
¹⁹ Rep. Yvette Clarke, H.R.5586 – 118th Congress (2023-2024): DEEPFAKES Accountability Act, (2023), https://www.congress.gov/bill/118th-congress/house-bill/5586 (last visited Jun 12, 2024).
²⁰ Id.
²¹ Id.
²² Id.
²³ Id.
²⁴ Id.
²⁵ SB0149 – Artificial Intelligence Policy Act, Utah State Legislature (2024), https://le.utah.gov/~2024/bills/static/SB0149.html (last visited Jun 12, 2024).
²⁶ Id.
²⁷ Id.
²⁸ Id.
²⁹ House Bill 182, 56th Legislature – State of New Mexico – Second Session, 2024 (2024), https://www.nmlegis.gov/Sessions/24%20Regular/bills/house/HB0182JCS.pdf (last visited Jun 12, 2024).
³⁰ HB 919, Florida House of Representatives (2024), https://www.flsenate.gov/Session/Bill/2024/919/BillText/Filed/PDF (last visited Jun 12, 2024).
³¹ Jeffrey Fleishman, Will Hackers, Trolls and AI Deepfakes Upset the 2024 Election?, Los Angeles Times (2024), https://www.latimes.com/politics/story/2024-04-30/will-a-i-deepfake-videos-and-robocalls-threaten-the-2024-election (last visited Jun 12, 2024).
³² Id.
³³ Roy Rochlin, New York Law Revisions Will Make AI-Generated Political Ads Less Risky For Broadcasters., Insideradio.com (2024), https://www.insideradio.com/free/new-york-law-revisions-will-make-ai-generated-political-ads-less-risky-for-broadcasters/article_15ec7536-2889-11ef-b7d7-b702a03e4118.html (last visited Jun 12, 2024).
³⁴ Rob Harkavy, New York Clamps down on the Use of AI in Political Ad Campaigns, GLI (May 9, 2024), https://www.globallegalinsights.com/news/new-york-clamps-down-on-the-use-of-ai-in-political-ad-campaigns/ (last visited Jun 12, 2024).