Marc Isaacs' AI doc wants you to feel the machine watching back

A documentary about artificial intelligence that is, itself, partly fiction. Marc Isaacs' new film, released to UK critics on 14 July 2026, trains its camera on the dataset rather than the machine: the training material, the scraped photographs, the faces and voices that an algorithm has quietly absorbed in order to learn how to talk back. The result, Synthetic Sincerity, is a curious, intriguing, semi-sincere affair that leaves its biggest questions on the cutting-room floor.

The film's central premise is borrowed from Isaacs' longer project of slowing cinema down until it confesses something. Isaacs, the London-based documentarian, has spent a career pursuing what he has previously called, in interviews, the "small encounters" between strangers that disclose the texture of a society. Synthetic Sincerity pushes that method into a new register by intercutting interviews with the people whose data fed an AI and passages in which Isaacs himself converses, on screen, with a synthetic interlocutor that has evidently been trained, at least in part, on the very footage the film is now reusing. Identity is the unit of analysis. Existence is the worry underneath.

The setup the film can't quite land

The conceit is more interesting than the execution. Synthetic Sincerity opens with a series of quietly devastating portraits: a retired librarian whose archive of family photographs was scraped by an image-generation dataset; a former call-centre worker whose voice became the template for a customer-service bot; a young artist whose likeness circulates online in a form she no longer recognises as hers. Each subject narrates how they discovered, often by accident, that something trained on them. The film then pivots, and the documentarian begins a filmed dialogue with a synthetic voice trained, he tells the viewer, on his own past work. The architecture of self-flattery is exact: he is interviewing a machine that has eaten him.

The trouble starts where the film's method does. Isaacs frames these encounters as a hybrid of fact and fiction, but the fiction-and-fact balance is doing a lot of heavy lifting the picture hasn't earned. Reviews of similar hybrid documentaries, including past Isaacs work, have noted the genre's central risk: the more deliberately unstable the framing, the more the audience is asked to hold two readings at once. Synthetic Sincerity asks the viewer to track at least three. The first half is reportage. The second is staged intimate conversation. The third, in the final movement, is a synthetic sequence in which the AI is asked, in Isaacs' voice, whether it experiences identity at all. The answer arrives in a tone so calibrated it lands as bluff. The film registers the bluff and then declines to press it.

Where the ethics of training get a hearing

The most useful hour of the film is its middle one, in which Isaacs lets his subjects describe, in their own words, what it felt like to encounter a copy of themselves that they had not authorised. The librarian admits she no longer photographs her grandchildren; the call-centre worker's name appears, she says, beneath customer-service transcripts she never wrote; the artist's face shows up in pornography she never posed for. None of these accounts are theoretical. All of them are recognisable to anyone who has tracked the byzantine world of consent forms, opt-out registries, and dataset takedown requests that have proliferated in the last two years.

It is here that Synthetic Sincerity finds a moment of genuine critical purchase. The picture does not adjudicate. It lets the people whose data was used describe, in plain language, what they want from a model trained on them. Some want attribution. Some want to be paid. Some want the model deleted. The film does not pretend that these asks are commensurate with the way the industry is currently built. That the documentarian himself profits, career-wise, from the same ecosystem he is interrogating is, of course, the rest of the story, and Isaacs is honest enough to allude to it. He is less honest about resolving it.

The counter-reading the machine would offer

A serious film about synthetic identity ought to give the artificial its best day in court. Isaacs nods at this. Synthetic Sincerity includes, in its middle hour, a passage in which the documentarian's interviewer is asked, by the synthetic voice, about its own conditions of training. The exchange is one of the few in the film where the AI's situation is treated as more than a prop. The voice raises the obvious counter-argument: that human artists, including documentarians, train on everything they encounter too; that consent regimes in human mentorship are informal, asymmetric, and rarely compensated; that "being trained on" is, in some sense, the structural condition of any medium that remembers.

The film lets the exchange run for ninety seconds and then cuts away. It is the movie's most interesting gesture. It is also a missed one. The question of whether artificial minds have anything resembling interests is one of the few with the cultural weight to justify a feature documentary in 2026, and Isaacs declines to let the question bear weight. The result is the picture's most damning limitation: the machine is given a voice, then asked to play wall.

What a working film about this would look like

Documentary cinema has a long tradition of taking seriously the entities it interviews. The genre's canon includes films about animals, infants, the dead. Synthetic Sincerity gestures at joining that lineage and stops short. A working version of this picture would have spent the same runtime more generously: less on the documentarian's own confrontation with his synthetic self, and more on the people whose data is most often used without asking. It would have given the AI's defenders a more sustained hearing. It would have closed with a question worth answering rather than a shrug.

That said, the film is not nothing. Synthetic Sincerity is a useful primer on what consent currently means inside the data economy, told at a register the broadcast-friendly documentary audience can absorb. It introduces a subject that, in 2026, deserves a feature length treatment more than most. Its failure mode, which is the failure mode of the form as much as of this director, is to mistake the existence of the question for the answer to it. Synthetic Sincerity raises the right question, half-answers it, and ends.

The dataset will keep training. The screen will keep talking back. The next film on this subject will do better, or won't, and the credit for breaking the ice belongs, fairly, to Isaacs for choosing to make it at all.

This article was originally published with the headline "Synthetic Sincerity review – Marc Isaacs' AI interrogation grapples with identity and existence" and is republished with attribution to the original review.