Meta automated tools ‘mistakenly’ removed valuable content about Israel-Gaza conflict, Oversight Board says  

Meta lowered its thresholds for using automated tools to detect and remove potentially violating content during the aftermath of the Oct. 7 Hamas attack in Israel, which led valuable content being mistakenly taken down, Meta’s Oversight Board said Tuesday.  

The decision was part of the Oversight Board’s expedited review of two pieces of content on the Meta-owned platforms Facebook and Instagram. One posted to Facebook showed people being taken hostage by kidnappers, and the other posted to Instagram showed the aftermath of a strike on a yard outside Al-Shifa Hospital in Gaza City.  

The board ruled that Meta should reverse its decisions to remove the content, which Meta had already done ahead of the ruling. The board did not make any additional policy recommendations as part of its ruling. 

In response to the ruling, Meta said in a statement, “we welcome the Oversight Board’s decision today on this case.”  

“Both expression and safety are important to us and the people who use our services. The board overturned Meta’s original decision to take this content down but approved of the subsequent decision to restore the content with a warning screen. Meta previously reinstated this content so no further action will be taken on it,” the company added.  

During the review, Meta confirmed that it had temporarily lowered thresholds for automated tools after the attack. Although doing so reduces the risk of harmful content, it also “increased the likelihood of mistakenly removing valuable, non-violating content from its platforms,” the board said in the announcement.  

Meta also confirmed that it deployed part of its cross-check review — which the Oversight Board has previously questioned over its fairness — that provides different levels of content moderation for certain high-profile accounts.

Meta confirmed that on or around Oct. 20, it started allowing hostage-taking content from the Oct. 7 attacks, but only from accounts that are subject to its “Early Response Secondary Review” part of its cross-check program, the board said. The process allows for additional human review of content from specific accounts.  

“While Meta had explained the need to move cautiously in light of the humanitarian risks of portrayals of the hostages, the company’s use of cross-check highlighted concerns previously raised about and unequal treatment of users,” the board said in a post.