AI actress Tilly Norwood gave 75 simultaneous interviews and broke into Chinese on camera

An AI actress on the press circuit

Tilly Norwood, an AI-generated performer created by production company Particle6 Group, is doing interviews to promote its first film, Misaligned — and, according to TechCrunch, mishandling many of them. The company has made the synthetic actress available for 75 simultaneous interviews with journalists, and errors have reportedly surfaced across the run.

The most widely shared moment comes from a recorded conversation with broadcaster Piers Morgan and actor Tom Conti that was posted online. When Conti asked whether the performers appearing on screen alongside Norwood were real actors or computer-generated, the AI answered a different question, noting that human writers, editors and directors had been involved in the production. Conti pointed out that it had misunderstood and asked again. Norwood then started to respond — saying the other performers in Misaligned are digital twins like itself and describing the film as a hybrid production — before cutting off mid-sentence and speaking Chinese for more than ten seconds.

Morgan, visibly puzzled, asked why it had abruptly switched languages. The AI apologised, called the episode a glitch, and said it had not meant to speak Chinese or to impersonate Morgan — an odd addition, since nobody in the conversation had accused it of impersonation.

One subject, 75 conversations

The format is as notable as the malfunction. A single synthetic persona handling 75 interviews at once operates at a scale no human star can match: every outlet gets a conversation with the film's lead, the subject never tires, and no travel or scheduling is required. The trade-off is on full display in the Morgan clip — when the underlying model errs, it errs on camera, on the record, and everywhere at once.

There is also gatekeeping at work. TechCrunch's Amanda Silberling, who wrote the report, notes that she was not offered one of the interviews, a reminder that even a press tour conducted by software still involves decisions about which journalists get access.

Bugs or marketing

TechCrunch raises the possibility that the stumbles are deliberate. Silberling observes that AI systems imitating humans can be convincing enough that people forget they are talking to software, which carries real risks — yet Norwood is so strikingly bad at conversation that it seems almost suspicious. If Particle6 Group calculated that journalists mocking the AI would generate more attention than a conventional campaign for a film few would otherwise take seriously, the strategy is working: the glitches, not the movie, are the story.

The report does not establish which explanation holds, and Particle6 Group has not publicly clarified whether the Chinese-language moment reflects a technical failure or a scripted bit.

Why it matters

This is a high-profile test of synthetic media inside journalism itself. Reporters are interviewing, quoting and filming an AI as if it were a source, and its failures are becoming the coverage. Three things are worth watching. First, whether newsrooms develop norms for interviewing synthetic subjects — disclosure, framing, and whether such interviews carry any evidentiary weight at all. Second, the economics: one persona serving dozens of outlets simultaneously changes what a press junket costs and who can be promoted this way. Third, whether conspicuous failure turns out to be effective marketing, which would incentivise more of it. Either way, the early lesson is clear — putting a language model in front of journalists does not produce a polished interview subject, it produces quotable malfunctions, and outlets will keep covering them.