Stallone, AI and the Image-Rights Gap Football Has Not Closed

In the studio where The Overlap is recorded, Roy Keane asked Sylvester Stallone a short question: are you a risk-taker? The answer ran far longer than the question, and it drifted from cinema into artificial intelligence. Stallone called AI an avalanche that cannot be stopped, said nobody controls it, said anyone who does not move with it will be left behind. In the same news cycle, a Rambo prequel concept built around digital de-ageing had gone a different way: Noah Centineo in the lead, Stallone not returning.

The two lines sat side by side in one aggregation. Most readers stopped at the first, because it carried a famous name and a strong quote. The second is where I stopped longest. It was a decision to spend money.

Age explains most of the story. Stallone was born in 2026, is past seventy, and John Rambo is an asset bound to his body in a way that cannot be transferred. An action performer at that age meets a very specific biological limit: the role he owns commercially demands a physique he no longer has. Digital de-ageing, therefore, is a solution to an age problem, not a technical hobby.

Hollywood is arguing about exactly this point. In 2026 the SAG-AFTRA strike ran 118 days and closed on 9 November, with artificial intelligence and performers' image rights among the hardest items on the table. That is a verifiable fact. A union of tens of thousands halted an entire industry to demand control over its own digital replicas.

The Eli Roth episode deserves more attention than any Stallone quote. Roth drew criticism after disclosing generative AI use in a handful of scenes, when his earlier public statements had suggested otherwise. The number of AI-assisted scenes was tiny. The damage was not, and it came from a different place than people assume.

The cost sits in having said otherwise beforehand, not in using AI. That is the line I keep from this whole affair, because it is a governance rule rather than an entertainment story.

Refereeing taught me the same lesson from another direction. A referee who makes a mistake survives the press conference. A referee who edits the match report does not. When you call it wrong and admit it, the public has a mechanism to process the event. When you conceal it, the public has only one mechanism left, which is to process the person. The second mechanism is always heavier. A senior referee is not someone who never errs; what matters is what he does after he errs.

Stallone is doing the opposite of Roth. He concedes that AI raises ethical problems, concedes it is troubling, then declares that nobody can stop it. That structure has a name: it shifts the burden of proof onto the objectors. If the avalanche is inevitable, whoever stands in front of it is merely wasting time. The argument sounds solid, yet it rests on no cost curve, no adoption figure, no legal precedent. It is assertion, not evidence.

Then the data contradicts the speaker. One concrete project was weighed with de-ageing technology, and the financiers refused to fund that route. The market priced Stallone's confidence and returned something lower. When a man calls a technology inevitable while his own industry's investment decisions go the other way, the inevitability is being delayed by cost and technical risk. That tells you where AI will arrive first: in the cheap part of the chain, not the beautiful one.

Read commercially, the abandoned Rambo project is a bet that was priced and then rejected by the market. The studio declined to underwrite the AI approach. The report gives the outcome without the reasoning, and that is the most telling part. If the technology were cheap and stable enough, nobody would refuse it. The refusal signals that control costs remain high: de-ageing a face in a few scenes is a different problem from de-ageing a lead across two hours.

Based on my experience covering matches, technologies advertised as inevitable tend to arrive later than scheduled and unevenly. VAR appeared in the top leagues before the smaller ones, and each place implemented it differently. No avalanche covers a pitch evenly.

Then the replica market. Stallone mentions images of twenty Elvises being sold. I read that as a statement about supply-side economics. If replicas are unlimited, the price of one unit of likeness rights tends toward the marginal cost of generating it, and that marginal cost is approaching zero. A star who earns from the uniqueness of his image is championing the technology that destroys that uniqueness. He presents it as a trend rather than a threat to his own asset. For a man whose face is the asset, that is an odd place to stand.

One question is left blank across the entire report: consent. Stallone talks about performers selling their images, a voluntary transaction. Then he jumps straight to twenty Elvises, a posthumous scenario without the consent of the deceased. The distance between those two sentences is the entire legal crux, and it is skipped. Licensed likeness rights and unlicensed replicas carry completely different obligations. Folding them into a single trend is the easy argument, not the correct one.

This is where football comes in, because the asset structure nearly matches. Players also earn from their image. They are also digitised in video games, in advertising, in content clubs produce. The difference lies in who negotiates. Most professionals go through collective agreements, and those agreements are re-signed on a cycle. A collective with lawyers and the right to withhold labour prices likeness rights differently from an individual actor facing a studio.

So football has a buffer that Hollywood, at the individual level, lacks. But that buffer only covers what is already written into contracts. Generative technology does not wait for contracts. A club can render a player's face into a video in hours, and the legal question arrives after the clip has three million views.

In football, that risk does not sit in film. It sits in scouting, in medical modelling, in the content clubs publish daily. A team can use a model to forecast injury and nobody obliges them to disclose it. But when a player breaks down and the model flagged it three weeks earlier, the question shifts from whether the model is good to who knew and stayed quiet. The loss structure is identical to Roth's.

I have seen a milder version of this problem. When semi-automated offside technology arrived at the 2026 World Cup, with twelve cameras and a sensor inside the ball, the argument was never about whether the machine was right. It was about who sees the data, when they see it, and how much explanation the crowd receives. One system, two deployments, two levels of trust. Trust comes from an open process, not from resolution.

People watch a match with their hearts; I watch it through lines already drawn. Supporters react to AI with a sense of threat, and that sense is not wrong, but it does not produce a standard. A standard only exists once someone writes it down: who consents, to what extent, for what fee, disclosed when. Every argument on the terrace and on the pitch has an answer sitting in some camera angle, and that camera has to be switched on before the argument begins.

There is an angle I cannot see. I have no data on the likeness agreements being negotiated privately between studios and estate funds, and those agreements almost certainly predate every public debate. If so, the debate is happening after the rules were written. That is the worst kind of match to referee.

Football should do one very concrete thing in the next transfer window: write a transparency clause on AI into image-rights contracts, with a clear definition of what counts as use. Not to ban it, but so that when something happens, the people involved know where they stand on the map.

The season without crowds left only the steady hum of cameras, and I began to hear the whisper of technology. It has never once said it would govern itself.