Meta Reportedly Refuses To Remove AI Videos of Child Abuse Showing a 'Baby Roasting in an Oven'

Meta is facing renewed scrutiny after Futurism reported that AI child abuse videos remained on Facebook after the publication flagged accounts containing them to the company.

Futurism said it sent Meta 19 accounts containing graphic AI-generated videos depicting children being abused, and the accounts remained online more than 24 hours later, and that Meta had not responded to its latest request.

The report raises questions about how effectively platforms detect AI-generated material depicting violent abuse and highlights the difficulty of distinguishing fictional or synthetic content from material that violates platform safety rules.

Meta published a statement on 7 October saying it investigates potential child-exploitation content and removes material that violates its policies. The company also said it had deployed additional AI tools to strengthen its detection systems.

Futurism Reports More AI Child Abuse Videos

Futurism's latest investigation follows an earlier report published on 9 September about graphic AI-generated videos appearing on Facebook. The publication previously reported videos depicting children being beaten, burned, restrained and subjected to other forms of violence.

According to Futurism, Meta removed some accounts or material identified in its earlier reporting. However, the publication said it later restored one account it had previously flagged before it disappeared again.

A Meta spokesperson previously told Futurism, 'Just wanted to let you know that we're looking into this.' The spokesperson added, 'I'm not sure of an exact timeline, but I will let you know when I learn more.'

Futurism reported that the account subsequently disappeared again. The publication said Meta did not answer its follow-up questions about the restoration.

For its latest investigation, Futurism said it sent Meta 19 additional accounts containing similar material. The publication reported that the accounts remained accessible more than 24 hours after it submitted them for review.

The videos described by Futurism include a baby being placed in an oven, a toddler being pushed down stairs, and children being assaulted with household objects. The publication said it chose not to publish the videos themselves because of their disturbing nature.

One described clip shows a girl locked in a pigsty saying, 'Grandpa, I'm so hungry.' Futurism also described videos depicting Black children being assaulted and verbally abused by people presented as teachers.

Some of the videos reportedly end with an adult intervening to rescue the child. Futurism said Meta had previously pointed to such endings when explaining why some material had remained available.

Futurism argued that detailed captions accompanying some videos could provide additional signals for automated moderation systems. However, whether individual captions can reliably identify prohibited material is a technical question that the examples alone cannot establish.

Meta Gives Its Side as Moderation Questions Grow

Meta published its own child-safety statement on 7 October, the same day as Futurism's latest report. The company said it investigates potential child-exploitation content when it becomes aware of it.

Meta said it had actioned 33.2 million pieces of child sexual exploitation content across Facebook and Instagram between January and June 2026. It said it identified and addressed more than 97% proactively, before users reported it.

The company also said it had introduced additional artificial-intelligence measures. These include large-language-model detection, AI-driven sweeps and systems designed to identify repeat offenders.

Those figures concern child sexual exploitation content and should not be treated as a direct measurement of the violent AI-generated videos described by Futurism. They nevertheless provide important context about Meta's stated approach to child-safety enforcement.

Meta's public response does not specifically address Futurism's 19 accounts in detail. Futurism's account of what happened to those accounts therefore remains the basis for the specific allegations concerning the latest test.

The report also follows a separate WIRED investigation into hundreds of advertisements containing child sexual abuse material on Meta's platforms. WIRED reported that some of those advertisements included images of real children.

That investigation involved sexual abuse material and advertising, making it distinct from Futurism's report about graphic AI-generated videos depicting physical violence against children. WIRED reported that Meta removed the advertisements and said it did not tolerate child exploitation.

Meta generally requires users to be at least 13 to use Facebook and Instagram. The company has also said it uses artificial intelligence and other systems to identify accounts that may belong to underage users.

The latest reports nevertheless raise broader questions about large social-media platforms' ability to detect harmful synthetic content. They also show why claims about moderation failures should be assessed against both independent reporting and companies' documented enforcement practices.

IBTimes UK has not independently reviewed the videos or the 19 accounts Futurism describes. The specific claims about those accounts should therefore be understood as allegations reported by Futurism, not independently verified findings.

The case highlights a growing moderation challenge as increasingly realistic AI-generated material appears on social-media platforms. It also underscores the importance of distinguishing verified platform actions from allegations about individual moderation decisions.