TikTok whistleblower warns AI moderation risks a 'generation’s mental health' - as millions still able to watch Perez Hilton self-harm livestream

Hilton, 48, was rushed to hospital earlier this month after he live-streamed himself self-harming on TikTok while viewers begged him to stop.

The highly graphic livestream remained online for around 15 minutes and was reported thousands of times before it was taken down by TikTok, with users criticising the social media platform for the slow response.

Asked why it had taken the platform so long to remove the livestream, a US TikTok spokesperson said the video had been flagged by its AI moderation system within minutes.

Read more: Perez Hilton's family says blogger 'can communicate' after TikTok self-harm livestream

Read more: Harry and Meghan slam Meta over $17bn child safety settlement, claiming payout a bid to dodge 'accountability'

The livestream was then sent to a human moderator for review, who was presented with an “error” message while trying to take it down, the spokesperson told Vanity Fair .

But three weeks later, hundreds of versions of the video continue to circulate on TikTok - with a whistleblower warning LBC of "dramatic consequences on mental health for an entire generation of young people".

Lynda Ouazar, who worked at the platform as a content moderator between 2022 and 2025, told LBC that AI was “clearly not ready” to replace human moderator teams.

Lynda was one of 400 content moderators who lost her job in the round of redundancies to TikTok’s London safety teams in December 2025.

When she worked at TikTok, human moderation teams were "working 24 hours".

“There were always humans who would be really easily able to see that this content should not be on the platform and remove it," she said.

But she added with AI "it's completely messed up...AI cannot understand if someone is holding a pen in their mouth or if it's a cigarette.

"AI cannot differentiate whether someone on TikTok is holding a knife because they want to commit a crime, or if they are holding a knife because they are cooking.”

As a "test", she searched for the Perez Hilton livestream on TikTok and was unsurprised to find it had been saved and re-uploaded "hundreds" of times.

When she filed a report about one of the reposted clips of Hilton’s livestream on 19 August, Lynda received a reply an hour later informing her the video had been reviewed but “no violations” of the platform’s Community Guidelines had been identified.

She added she “regularly” reports videos of pornography, child nudity and swearing which “don’t get taken down” by the platform.

"The reason why Perez's video went viral is only because he is a celebrity, but there are hundreds if not thousands of other videos like that or worse that people do not know about. 

"I see them myself, I report them and they’re still there," she said.

While working for the company, Lynda was involved in training up AI systems, which at the time was framed as a way of helping to shield human moderators from being exposed to the most harmful content on the internet.

As part of that process, footage would be moderated by both human teams and by AI, with TikTok then comparing how the AI tools applied the platform’s policies on violations compared to the human teams.

“We were reviewing the results every week and AI was getting better, but very slowly. Our response was the same each time: the AI is not ready," she said.

The company also more recently laid off 300 jobs from its trust and safety teams in Ireland, and also recently announced it would be laying off all 250 of its content moderators in Nashville and closing the office entirely.

Lunda alleges the mass lay-offs in the London office followed moves by staff to unionise in response to poor working conditions, low wages and a lack of psychiatric support for staff exposed to highly traumatic content online.

She is currently pursuing legal action against TikTok along with three of her former colleagues over claims of bullying, harassment and union busting.

Martha Dark, co-founder of Foxglove, said: "TikTok was quick to pin any kind of moderating success on the AI.

"But I noticed they blamed the reason the Perez Hilton video stayed online for so long on an individual content moderator."

She said: “For a long time, all of the social media platforms have been telling the world how effective their AI is at moderating content and that it has nothing to do with the human reviewers.

“And the reality, as someone who has worked with content moderators for the last six or seven years, is that social media content moderation will always, to an extent, need to be done by a person, partly because the AI isn't anywhere near smart enough, but partly because there is some nuance and complication."

She said it was "unlikely" that AI would ever be able to entirely take over moderation of large social media platforms.

"I think it's a size issue. We have allowed these companies to become so big and so powerful that they are essentially impossible to moderate.

"There's so much content and so many users, and they're just not being forced to properly moderate it.

"And these companies aren't going to change on their own until they're forced to because of the financial investment that proper content moderation requires."

Last year, one of TikTok's Northern European chiefs was forced to defend the job cuts arguing that a “very high benchmark” had been set for user safety.

According to the firm, 85% of the content removed for violating its Community Guidelines is identified and taken down without requiring any human input.

Ali Law, TikTok's director of public policy and government affairs for Northern Europe, said that while in the past, AI moderators would have been able to identify a knife, newer models could determine the context - whether they are used in a cooking video or in a violent encounter, for example - something directly contradicted by Lynda.

He said: "We also make sure the changes are introduced on a gradual basis with human oversight so that if there isn't a level of delivery in line with what we expect, we can address that."

It followed criticism from the Chair of the Science, Innovation and Technology Committee, who accused TikTok of failing to supply evidence in support of its claim that job cuts increased the efficacy of moderation and said staffing cuts posed a “real risk to the lives of TikTok users”.

TikTok has been approached for comment.