Meta reports removing 20 covert AI-linked influence operations around 2024 elections

Meta published a review of its platforms during 2024’s national elections, held in more than a dozen countries with an estimated two billion voters combined, reporting that it had removed around 20 covert influence operations over the year — most originating in Russia, which the company said had accounted for 39 of the networks it has disrupted since it began tracking them in 2017, ahead of Iran’s 31 and China’s 11.

On generative AI specifically, the company reported the opposite of what had been widely feared going into the year: networks it took down used AI tools for incremental production gains rather than to expand their reach or evade detection, and AI-related content made up a small share of all election misinformation its fact-checking partners flagged. Meta president Nick Clegg was quoted saying generative AI had not proven “a particularly effective tool” for evading the company’s enforcement. Meta also disclosed that its own image generator rejected roughly 590,000 requests to depict US political figures in the month before Election Day.

These risks did not materialize in a significant way.

— Meta, “What We Saw on Our Platforms During 2024’s Global Elections”

Meta attributed this partly to a detection strategy built around behavioural signals — coordination patterns, fake account clusters — rather than on whether content was AI-generated, meaning enforcement did not depend on catching every synthetic image or text. Independent researchers and journalists had documented individual instances of AI-generated election disinformation during the year regardless, so the report described a company’s self-assessment of its own defences rather than an independent audit of the problem.