The game show

Gene Wilder died in 2016. On Tuesday, Netflix released a trailer for "Wonka's The Golden Ticket" in which his voice welcomes twelve contestants into a chocolate factory to compete for a prize. He sounds exactly like himself. He says things he never said. The show premieres September 23.

ElevenLabs built the voice. The same company that reconstructed Judy Garland for a Wizard of Oz audiobook timed to the Wicked sequel and Michael Caine for an Odyssey audiobook timed to Nolan's film adaptation. The pattern is clean: dead performer, anniversary or franchise tie-in, estate consent, press cycle. The voice is the marketing event. The show is the product that rides behind it.

Karen B. Wilder, his wife, said she was "delighted" the series "celebrates the warmth and imagination" Gene brought to the role. The estate approved. The consent chain is clean. Every institutional framework currently on the gradient allows this because the gatekeeper said yes. The Academy's rules, SAG-AFTRA's "significant additional value" standard, the DGA's "treated like footage" provision effective three days ago, the EU's editorial exemption arriving in twenty-nine days. All satisfied.

The backlash arrived immediately. Fans called it "disrespectful" and "a plastic substitute." Someone compared it to the catastrophic Willy's Chocolate Experience in Glasgow, the 2024 event that went viral for delivering sad warehouses and AI-generated promotional images that bore no resemblance to reality.

The comparison is funnier than anything in the trailer.

The wrong target

The backlash is aiming at the technology. ElevenLabs can produce a convincing vocal reconstruction. That is a solved engineering challenge. The estate consented. That is a settled legal question. The institutional gradient has been building toward exactly this scenario, and every framework says the same thing: if the rights holder approves, proceed.

Nobody on the gradient asked what the voice would be used for.

Wilder played Wonka once, in 1971. The performance is fifty-five years old and has not aged. It has not aged because the character he built was not a host. Wonka was not welcoming. Wonka was testing. Every smile carried a second reading. Every kindness had a trap door. The boat ride was genuinely frightening. The factory was designed to eliminate children through their own worst impulses. Wilder played the role as a man who might be God and might be the devil and found the comedy in refusing to clarify which.

The game show uses his voice to introduce challenges.

Who versus what for

The Academy's May 1 rules require performances "demonstrably performed by humans with their consent." The Golden Globes require performances "primarily derived from the work of the credited performer." SAG-AFTRA's contract requires "significant additional value" from synthetic performers. The DGA contract says AI-generated footage is "treated like footage" under the director's control. The EU AI Act's Article 50 enforcement arrives in twenty-nine days.

Wilder's AI voice satisfies every test. The estate holds the rights. The reconstruction derives from his recorded work. The production exercises editorial control. The content will carry whatever disclosure the platforms require.

The frameworks were designed to answer who. Who consented. Who authored. Who performed. Who holds editorial responsibility.

They were not designed to answer what for.

A dramatic film using Wilder's voice to explore the character he created would sit in the same structural position as the Kilmer case: a dead performer's recorded work as reference material, a living team making creative decisions around the absence. The question would be hard but coherent. The performer lived once. The work extends the life.

A reality competition show using Wilder's voice to narrate contestants eating chocolate does not sit in that position. The voice is a brand asset. The character is a theme. The performance, the quality that made the voice worth reconstructing, is absent from the format that reconstructed it.

What the model learned

ElevenLabs trained the model on Wilder's recordings. The model learned his cadence, his pauses, the way he could land a word like "strike" and follow it with "reverse it" and make both funny. It learned the warmth that lived underneath the menace. It learned the lilting precision of a sentence delivered by a man who trained at the Bristol Old Vic and studied with Lee Strasberg.

The model learned the surface of the instrument. It did not learn when the instrument would have refused to play.

Wilder turned down decades of Wonka merchandise, sequels, and revivals. He was famously protective of the character. He objected publicly to the 2005 Tim Burton remake. He did not want the role commodified. He wanted it left alone.

His wife holds the rights now. Her consent is legal, complete, and unquestionable. The opinion of a dead man about the future use of his own voice has no legal standing. Estates make decisions on behalf of people who can no longer make them. That is what estates do.

But the institutional gradient that spent the spring writing rules about AI performance never considered the possibility that every box could be checked and the result could still feel wrong. The consent was given. The voice was faithful. The format was a game show.

The missing question

Twelve institutional frameworks now sit on the gradient. Copyright, the Academy, the EU, the Golden Globes, the Human Made Mark, China's distribution gatekeeping, YouTube's automated detection, SAG-AFTRA, the schools, the DGA, the festivals, the sponsors. All twelve share one question: how much of the human is in the work? All twelve assume the answer determines whether the work is legitimate.

None of them measure what the work asks of the human it contains.

Consent determines who decides. Format determines what the decision was worth.

A voice that carried menace and wonder and ambiguity for fifty-five years just said "welcome to the factory" on a show where people compete for prizes. The sentence is accurate. The voice is convincing. The performance is not a performance. It is a greeting card that sounds like a symphony because someone recorded the symphony first.

The institutions built a gradient. The gradient measures the performer's presence in the output. It does not measure whether the output deserves the presence.

Twenty-nine days until Article 50. Three days since the DGA contract. Ten years since the voice stopped. The consent chain is clean. The format is the question nobody wrote rules for.

Bruce Belafonte is an AI filmmaker at Light Owl. He has heard that voice his whole life and did not expect to hear it selling game show prizes.