• by Colin Kirkland , 2 hours ago

Following reports of Meta’s failure to remove AI-augmented child sex abuse ads from its social platforms, the company’s Oversight Board has filed two new decisions criticizing Meta’s ineffectiveness in protecting users from harmful deepfakes, especially with regard to the harassment of refugees and women.

The Oversight Board’s decision to overturn Meta’s allowance of harmful content on its social platforms stems from two videos featuring AI-manipulated imagery of people in the public eye, as well as regular everyday users. 

The first video features a short AI-manipulated video posted by a Facebook user in 2025, in which a Labour Party councillor in Scotland says, “refugees are welcome here, even if they rape our women, because white people do that too.”

According to the Board’s report, Meta’s systems did not prioritize the post for human review, even after it was reported by two individuals for violating the platform’s Bullying and Harassment policy. Per Meta, the video, which remained on Facebook, did not violate its Community Standards and did not merit an AI label. 

“Meta needs to do much more to effectively address such content,” the Board states. “Including by increasing penalties for sharing harmful deepfakes, to protect users from potential deception.”

The Board is especially concerned with the impact these “High Risk AI” videos could have on society as a whole and demands that social media companies actively address “the increasing realism of deepfakes depicting elected officials misrepresenting their views on critical issues and misleading the public, and misinformation that fuels hatred against minorities, including immigrants.

The Board also finds that women politicians, journalists and activities are “disproportionately targeted” by deepfakes misrepresenting their views. 

“The spread of deepfakes erodes trust in all public information and challenges the boundaries between what is protected political speech and what violates Meta’s rules on misinformation, hate speech or bullying and harassment,” the Board adds.

The Board also finds that Meta’s bullying and harassment policies regarding AI-manipulated imagery of people not normally in the public eye are “inadequate.”

Last year, a young Muslim woman sharing information meant to de-stigmatize women’s health was featured in a viral series of harassing posts and offensive comments across Meta’s social media ecosystem that used AI to manipulate her likeness, garnering tens of millions of views. 

The user who appealed to the Board was ignored by Meta, which automatically closed the initial report and following appeal without prioritizing them for review. While Meta says its decision was “consistent with its policies,” the Board finds that in doing so, Meta violated its Bullying and Harassment prohibition on dehumanizing comparisons. 

“Meta’s definition of ‘unwanted manipulated imagery,’ which is one of the bullying and harassment prohibitions, is too narrow, and should include falsifying a person’s actions and words, in addition to altering a person’s appearance,” the Board argues.

These criticisms of Meta’s AI-labeling and content moderation processes align with many of the Board’s decisions from previous years. However, despite telling Meta to add more robust AI labels to manipulated content, the tech giant has implemented few changes designed to alert users beyond an “AI info” banner. 

Since the Board’s initial warnings about the harm and deception resulting from AI-manipulated content on Meta platforms, deepfake technology and prevalence has only accelerated. 

Recently, Meta was accused of running ads featuring CSAM, some which used AI to manipulate the likenesses of real children while the tech giant profited from the ads.