The Senate Judiciary Committee advanced the NO FAKES Act (S.1367) by voice vote on June 18, 2026, creating — for the first time under federal law — an intellectual property right over a person's voice and visual likeness. Platforms that fail to make a good-faith effort to remove unauthorized AI-generated replicas on notice face penalties of up to $750,000 per work. It is the most substantive federal deepfake legislation since the TAKE IT DOWN Act, signed by President Trump on May 19, 2025, which addressed nonconsensual intimate imagery but left commercial voice cloning and likeness exploitation entirely unregulated at the federal level.
The Harms Are Real
Before dissecting the bill's flaws, the case for federal action deserves fair hearing. Nonconsensual AI voice cloning has produced unauthorized commercial recordings mimicking real musicians. AI-generated replicas of deceased artists and public figures have appeared in advertisements and songs without estate consent. The state-level response has been voluminous but incoherent: as of April 2026, Ballotpedia tracked 47 states with some form of deepfake law — yet the coverage varies so widely that a replica lawful in one state may be actionable in the next. SAG-AFTRA gathered more than 16,000 member signatures supporting the bill, and the underlying complaint from creative workers is well-founded: generative AI has created real substitution pressure on licensing markets for voice and likeness. The TAKE IT DOWN Act established the notice-and-removal principle for intimate imagery; the NO FAKES Act extends that logic to the broader class of commercial exploitation. That extension is analytically coherent.
What the Bill Does
S.1367 grants every individual a federal IP right in their voice and visual likeness. Distribution of an unauthorized digital replica — an AI-generated voice or likeness in an audiovisual work or sound recording — without the subject's license is prohibited. That right persists after death and is transferable to heirs for up to 70 years, a term that mirrors copyright duration rather than the more modest terms in most state right-of-publicity statutes.
On receiving notice, platforms must make a good-faith effort to remove infringing content. The 2026 reintroduction added a counter-notification procedure modeled on the DMCA, allowing users to contest removals — but it requires a notarized physical signature, which is a meaningful barrier for ordinary people trying to defend commentary or satire from an unjustified takedown.
The bill includes exemptions for news, documentary, commentary, and parody — but those exemptions are qualified. News content is exempt only when the digital replica is the subject of or materially relevant to the story, a condition that makes routine editorial decisions potential legal exposure. Exemptions also extend to nonprofit libraries and accredited educational institutions for non-commercial uses.
The bill passed committee with 15 bipartisan co-sponsors including Sen. Chris Coons (D-Del.) and Sen. Marsha Blackburn (R-Tenn.), and has drawn support from OpenAI, Google, YouTube, TikTok, the Recording Industry Association of America, and the Motion Picture Association.
The Platform-Liability Problem
The bill's central flaw is its enforcement mechanism. Platforms that fail to make a "good-faith" takedown effort face up to $750,000 per work — but the bill provides no safe harbor for a platform that exercises genuine editorial judgment about whether flagged content is satire or parody and reaches the wrong answer. As EFF put it, the bill offers "no protection for a platform's judgment about an often difficult question — whether a particular piece of content is satire, parody, commentary, or news."
This creates a textbook chilling incentive: remove first, review never. The DMCA's notice-and-takedown system has a 25-year track record of over-removal even at its existing scope; the NO FAKES Act compounds the problem by requiring platforms to implement staydown monitoring — detecting and blocking re-uploads of flagged material — rather than simply responding to individual notices. The Computer & Communications Industry Association estimated first-year compliance costs for a startup-stage company at $1.64 million, driven largely by content-fingerprinting infrastructure and notification processing.
Even senators who voted to advance the bill were uneasy. Sens. Ted Cruz (R-Texas), Mike Lee (R-Utah), and Eric Schmitt (R-Mo.) each raised First Amendment objections. Cruz cited AI-generated campaign parodies as speech deserving protection, not liability. The concern is not theoretical: the same notice-and-takedown mechanism that removes child sexual abuse material has also taken down political commentary, criticism of public figures, and documentary footage under copyright claims. Adding a new IP right with steeper per-work penalties and no editorial safe harbor is likely to make that problem worse.
EFF raised a further structural concern: because the bill creates a transferable property right in likeness and voice, studios and record labels are almost certain to require workers to sign it away in standard contracts. The legislation presented as protecting artists could, in practice, transfer enforcement control to the same industry incumbents who already hold dominant positions in licensing.
What Proportionate Regulation Looks Like
A federal deepfake protection law is genuinely needed. Forty-seven inconsistent state statutes cannot adequately address harms that cross state lines instantly. But the current bill's platform liability mechanism requires three targeted repairs before a Senate floor vote.
First, the safe-harbor standard should be objective and procedural, not outcome-based. If a platform follows a documented review process — including a meaningful counter-notification window — it should be shielded from the $750,000 penalty even if a court later determines the content was infringing. The DMCA model, for all its well-documented failures, at least follows this principle.
Second, the 70-year posthumous rights term is excessive and poorly calibrated. Most state right-of-publicity statutes set terms of 10–100 years, with the median near 40. A 70-year term — identical to works-made-for-hire under copyright — serves IP alignment with entertainment contracts more than it serves any privacy or dignity rationale.
Third, statutory damages should require some showing of commercial harm or intent, as several state deepfake laws do. Not every unauthorized AI replica inflicts the same injury as a commercial voice-clone for hire; the penalty structure should reflect that.
Conclusion
The NO FAKES Act's core protection is sound: federal law should prohibit commercial AI exploitation of voice and likeness without consent. But sound policy goals and sound enforcement mechanisms are different things. A $750,000-per-work platform penalty with no editorial safe harbor will produce a system that suppresses satire and criticism at scale — a cure with its own significant harms. The bill's bipartisan committee majority gives it genuine Senate floor momentum. Congress should use that momentum to fix the liability clause, not simply to pass the bill as written.