Zelda Williams to AI video makers: 'have some shame'
The family of Robin Williams has spent three years asking people to stop making synthetic versions of him. This time the ask came with a description of the business model behind it.
Zelda Williams went public on X on Monday night, after an AI-generated video presenting her father discussing conspiracy theories circulated widely enough that fans sent it to her. "That supposedly 'private video' of dad that I've been shown talking about conspiracies is clearly AI, and not even particularly convincing AI," she wrote. "Anyone who claims to be a fan who believes it clearly listened to his movies on mute because that voice is robotic and terrible." Then she turned to the person who made it: "A human made a robot to create it, and to you I say: leave him out of your delusional bulls--- and let him rest." She closed with the line that traveled: "Just because he's gone does not mean he's now your puppet. Have some shame."
The detail worth sitting with is that the video did not need to be good. Robin Williams died in 2014, so he cannot contradict a clip, and the people who pass one along are not reviewing it frame by frame. The payoff is not persuasion, it is volume — a recognizable face plus a rough approximation of a famous voice is enough to farm engagement. "Not even particularly convincing" and "still circulating" are not in tension; they are the same business model.
This is the family's third public escalation. In August, Williams' three children — Zak, Zelda and Cody — revived their father's dormant Instagram account after twelve years, framing it as a direct counterweight to what they called rampant AI abuse of his voice and likeness: "As technology evolves, we want this to be a safe, trusted place where the stories, photos, videos, and memories shared reflect his legacy with authenticity, warmth, and care." Before that, Zelda asked fans in October 2025 to stop sending her the videos at all, calling them gross and telling people to "stop doing this to him and to me, to everyone even, full stop," and in 2023, during the SAG-AFTRA strike, she described the recreations as disturbing and said she was not an impartial voice in the union's fight over AI. What is new is the target: not the audience that believes the clip, but the person who built it.
The legal ground underneath all of this has moved, unevenly. California's AB 1836 took effect on January 1, extending the state's post-mortem right of publicity to AI-generated digital replicas of deceased personalities. It sets a floor of the greater of $10,000 or actual damages, adds disgorgement of attributable profits plus attorneys' fees, and narrowed the old exemptions so that "expressive work" no longer covers nearly everything — leaving commentary, criticism, satire, parody, documentary use and fleeting or incidental use. A companion statute, AB 2602, makes digital-replica clauses in performers' contracts unenforceable unless the intended uses are described specifically and the performer had counsel or a union. Federally, the NO FAKES Act was reintroduced in May with bipartisan sponsors and would create a licensable property right in voice and visual likeness that survives death — the version estates have been asking for, and the one that would raise its own preemption questions about California's approach.
None of that solves what Zelda Williams is actually describing. A statutory right gives an estate a claim; it does not give a grieving daughter a fast way to get an anonymous account's video out of a recommendation feed. And the exemptions that survived in AB 1836 — commentary, criticism, satire, newsworthy use — are precisely the labels an engagement farmer reaches for first. Paul Walker's brother Cody has said he will not stand in for Brian O'Conner again, and that the decision belongs to Paul's daughter: the same instinct, that the family is the last line of consent, inside a system where the family holds no practical switch. Synthetic likeness is one of the few AI harms that lands on a private person rather than a corporation, and the remedies on offer were built for the corporation.
Courts have been circling the same principle from the other direction — we reported how China's top court set the rules of the road for AI deepfake lawsuits this month, and the mechanics of why these clips are cheap to make and hard to police are in AI 101 — What is a deepfake?.
What to watch: whether platforms offer estates a takedown path with the same speed as a copyright claim — that, not another statute, is the thing that would have kept Monday's video off her feed.
Should estates get a fast platform-side route to remove synthetic clips, or should that decision stay with the courts? Tell us in the comments.
