Meta's automated moderation mistakenly blocked political accounts multiple times during the U

Meta has replaced roughly half of its content-moderation requests with large language models as of June and aims for over 90% automation in some violation categories by year-end, claiming its models make 13% fewer errors and catch 10% more violations than humans. Brazil's Lula campaign filed a complaint with the Supreme Electoral Court on October 10 alleging Meta restricted its reach and mistranslated negative comments about rivals as supportive; the court ordered equal treatment and a daily fine of 500,000 reais (~$98,000), though no penalty has been ruled and prosecutors filed a civil suit on October 21. In the U.S., the No Kings movement's Instagram account was suspended for about 90 minutes on October 17, which Meta attributed to high-frequency reads via the Claude tool triggering automated safeguards, and Florida Democratic Senate candidate Angie Nixon's Spanish-language account was wrongly flagged as impersonation on October 27 and restored the next day. Meta has apologized and called the incidents technical errors, while its Oversight Board criticized repeated election-enforcement mistakes without finding political motive. Attorneys general from 13 states plus Washington, D.C., and members of Congress have demanded explanations, though their requests carry no legal force.