What regulatory steps are being taken to stop AI deepfakes in online advertising?
Executive summary
Governments and regulators are moving from observation to action on AI deepfakes, combining mandates to label and trace synthetic content with notice-and-takedown rules and criminal/civil remedies for nonconsensual imagery [1] [2] [3]. But the response is fragmented—EU-wide binding obligations and codes of practice sit alongside a patchwork of U.S. state laws and a recently passed federal statute, leaving enforcement, platform responsibility, and industry pushback unresolved [1] [2] [3].
1. EU: binding transparency rules plus a Code to operationalize labeling
Brussels has built one of the clearest regulatory backstops: the AI Act sets transparency obligations requiring providers to mark AI‑generated or manipulated content in machine‑readable formats and to label deepfakes used in public interest contexts, and the Commission published a draft Code of Practice to translate Article 50’s requirements into operational labeling standards ahead of full application in August 2026 [1] [4]. The draft Code is on a fast schedule—feedback closed in January and a final Code expected by mid‑2026—to give industry practical steps before the law’s obligations take effect [1] [4].
2. United States: federal notice-and-takedown plus dozens of state laws create a web of remedies
In the U.S., Congress and states have taken complementary but uneven approaches: the Take It Down Act mandates that covered platforms implement notice‑and‑takedown processes for intimate images and deepfakes by May 2026 and criminalizes threats to publish such content, while dozens of states have enacted their own deepfake or likeness‑protection laws addressing everything from nonconsensual intimate imagery to voice cloning [2] [5] [6]. Separate federal proposals—like civil remedies in the DEFIANCE Act for victims of nonconsensual sexually explicit deepfakes—have advanced in Congress and the Senate, showing federal interest in victim relief as well as platform obligations [3].
3. Labelling, metadata, and machine‑readability: the favored technical levers
A common technical demand across jurisdictions is labeling and traceable metadata: the EU’s Article 50 and draft Code require machine‑readable markings on AI content, New York’s ad disclosure law forces advertisers to label AI performers, and proposed U.S. rules (and some state bills) would require AI outputs to carry detectable provenance or metadata to aid detection and enforcement [1] [7] [3]. Policymakers view such technical markers as a tractable means to limit deception in advertising and political messaging, though exact standards and interoperability remain unsettled [1] [3].
4. Platform accountability, takedowns and the limits of law alone
Regulators are pressing platforms to act—notice‑and‑takedown systems, age verification, and swift removal windows are now law in some places—but courts and platforms have litigated the boundaries, citing Section 230 and free‑speech concerns that have already led judges to overturn or flag state statutes [3]. The Grok fallout on X/XAI has further highlighted how platform design choices and moderation failures can trigger regulatory scrutiny across jurisdictions, but governments still debate who bears ultimate responsibility when generative models produce unlawful outputs [8] [9].
5. Global patchwork and political friction: divergent models and hidden agendas
The regulatory landscape is a “compliance splinternet”: the EU’s rights‑and‑transparency‑first model contrasts with U.S. federal patchwork and state experimentation, while countries like South Korea pursue harsher liability for publishers of fabricated content—moves that U.S. officials have at times framed as censorship risks [10] [11]. Industry actors and some major investors are funding pushback against strict state rules, revealing an economic incentive to slow or shape regulation even as victims’ advocates press for stronger protections [3].
6. Enforcement challenges and what to watch next
Key enforcement gaps persist: technical standards for labeling, cross‑border takedown cooperation, and how to police synthetic content in political ads remain unresolved, and judges have already signaled constitutional and speech‑law hurdles for some state measures [1] [3] [8]. The immediate horizon includes implementation deadlines in 2026—platform takedown rules, the EU transparency regime going live, and continued state rollouts—making the coming year a stress test for whether law, tech, and civil society together can curb deepfakes in advertising and beyond [2] [1] [12].