Robin Williams’ Instagram account had been sitting untouched since 2014. This week it came back online, and the message his three children posted there was not a memorial. It was a boundary.
Zak, Zelda and Cody Williams wrote that they are “grateful to see new generations discovering his work while longtime fans continue to celebrate the joy he brought into their lives,” and then added the line that everyone noticed: “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.”
Nobody in that statement says the letters A and I. They did not need to.
The short version
- What happened: Robin Williams’ children reactivated his Instagram after 12 years dormant
- Why: a flood of AI generated video and voice recreations using his likeness
- The trigger point: a post marking what would have been his 75th birthday on July 21
- Zelda Williams has been asking people to stop since October 2025
- The legal reality: posthumous likeness rights in the US are a patchwork, and the federal bill meant to fix that has stalled
The account as a fence
There is a strategic logic to what the Williams family did, and it is worth spelling out because more estates are going to copy it.
An abandoned account is a vacuum. Search a dead celebrity’s name on any platform and you will find fan pages, tribute accounts, memorial pages and, increasingly, accounts that exist purely to pump out synthetic video of that person. None of them are the official one, because the official one stopped posting years ago.
Turning the real account back on changes the topology. It gives platforms a verified rights holder to route reports through, it gives fans an obvious canonical source, and it gives search engines a signal about which page is authoritative. It is fence building rather than litigation, and it is the only tool that works at the speed content actually spreads.
Zelda Williams has been saying this out loud for years
The Instagram statement is measured. Her earlier posts were not, and they are a much clearer read on what it feels like to be on the receiving end of this.
“Please, just stop sending me AI videos of Dad. Stop believing I wanna see it or that I’ll understand, I don’t and I won’t.”
“To watch the legacies of real people be condensed down to ‘this vaguely looks and sounds like them so that’s enough’, just so other people can churn out horrible TikTok slop puppeteering them is maddening.”
“You’re not making art, you’re making disgusting, over processed hotdogs out of the lives of human beings, out of the history of art and music, and then shoving them down someone else’s throat hoping they’ll give you a little thumbs up and like it. Gross.”
Zelda Williams, October 2025
Back in 2023, sharing a Screen Actors Guild campaign against AI recreations, she described them as “at their very best, a poor facsimile of greater people, but at their worst, a horrendous Frankensteinian monster, cobbled together from the worst bits of everything this industry is, instead of what it should stand for.”
The videos in question are not obscure. Circulating clips have shown Williams speaking to a younger version of his Good Will Hunting character, appearing at a fictional awards ceremony alongside Betty White, and delivering invented advice framed as coming from beyond the grave. Most were made with OpenAI’s Sora. Most were made by people who genuinely believed they were paying tribute.
How we got here
The dead celebrity loophole
When OpenAI launched Sora 2, it shipped with likeness controls for living people. You opt in, you decide who can use your face, and the system is supposed to block everyone else. It was a real improvement over the previous free for all.
Deceased public figures were handled differently. OpenAI has said plainly that it allows the generation of historical figures, which is a category with no clear edge. Historians are historical figures. So, apparently, is a comedian who died in 2014.
That policy collided with reality quickly when synthetic videos of Martin Luther King Jr began circulating. OpenAI paused MLK depictions and established a route for authorized representatives and estates to request that a figure’s likeness be excluded, saying that public figures and their families should ultimately have control over how their likeness is used.
Why an opt out process is not the same as a right
A request based system puts the burden on grieving families to monitor every model, every platform and every new tool, then file paperwork with each one. It scales for OpenAI. It does not scale for a family. And it covers exactly one company, while the open weight video models that anyone can run locally answer to nobody.
What the law actually offers
In the United States, the right to control your name, image and voice is called the right of publicity, and it is state law rather than federal. Whether it survives your death, for how long, and whether it covers AI generated replicas specifically, depends entirely on where you lived.
| Protection | Covers | Status |
|---|---|---|
| Tennessee ELVIS Act | Voice and likeness, explicitly including AI generated imitations | In force since 2024, state level only |
| California AB 1836 | Commercial use of digital replicas of deceased performers | In force, requires estate consent |
| New York digital replica law | Deceased performers, commercial exploitation | In force, 40 year post mortem term |
| Federal NO FAKES Act | A national right covering digital replicas, transferable after death and renewable up to 70 years | Introduced in the Senate, has not moved |
| Many other states | Nothing specific to digital replicas | No post mortem right at all in some cases |
Two more gaps make enforcement harder than the table suggests. Most of these statutes are written around commercial use, and a fan video with no ad revenue attached is not obviously commercial. And a US judgment is close to worthless against an anonymous uploader in a jurisdiction that never signed up to any of it.
That is the same wall people hit in other contexts where synthetic media outruns the statute, which we dug into when asking whether anyone can actually sue over an AI deepfake. The short answer keeps being: it depends who you are, where you are, and who made it.
Platforms are the real enforcement layer
Since the law is slow and fragmented, the practical control point is whoever hosts the video. Instagram, TikTok and YouTube can remove content in minutes, which no court can. That is exactly why an active, verified account is worth more to an estate than a legal theory.
Platforms have been building toward this, though unevenly. Labeling requirements for synthetic media, provenance metadata and detection systems have all rolled out in some form, with results that vary from useful to decorative. The struggle to reliably identify machine generated content at scale is the same one behind LinkedIn’s attempt to declare war on AI slop, and the honest assessment there applies here too. Detection is hard, and it gets harder every time the generators improve.
They have improved a lot. The gap between an obvious fake and a convincing one closed faster than almost anyone predicted, as anyone who has compared this year’s tools against last year’s can tell you from our rundown of the current AI video generators. A likeness that took a studio and a budget in 2023 now takes a text prompt and a minute.
The part that is not about technology
Most of the people making these videos are not trolls. They are fans. They loved this man’s work, they miss him, and a tool appeared that let them conjure one more moment of him. The intent is often affection.
The family’s answer to that is worth sitting with anyway. A recreation is not a memory. It is a guess about what someone might have said, generated by a system that has no idea who they were, presented in their face and voice. When enough of those guesses circulate, they start filling in the space where the real record used to be, and the people who actually knew him are the ones who have to watch it happen.
Their statement closed by saying he “continues to inspire us and delight people all over the world.” The account will be used for real photos, real clips and real stories. Which, given everything else on the internet with his face on it right now, is a modest thing to have to fight for.
What to watch next
- Whether the NO FAKES Act moves. A federal right is the only thing that would replace the current state by state patchwork, and it has been stuck since introduction.
- More estates doing the same thing. Reactivating dormant verified accounts is cheap, fast and effective. Expect a wave of it.
- Whether platforms build estate specific tools. Right now families report content the same way anyone does. A dedicated rights holder channel would change the math.
- Open weight video models. Every guardrail discussed here applies to hosted services. Models running on someone’s own hardware have no policy team to petition.
