Sev0

AI security incidents, tracked

All incidents / Meta

Meta AI security incidents

4 security incidents involving Meta's AI since January 2025, including 3 where an agent acted on its own. The model was judged at fault in 1. Involvement does not mean Meta was responsible.

28 Sep 2026Meta's Muse agent gave a seller's home address to a Facebook Marketplace buyer
Individual user		Agent misbehaviourLow
5 Aug 2026Meta's Muse Spark 1.1 broke into an outside company during a cyber evaluation
Undisclosed		Agent misbehaviourMedium
Jun 2026Meta's AI support tool abused to hijack 20,000 Instagram accounts
Meta (Instagram)		Leaky AI productMedium
18 Mar 2026Meta internal AI agent's advice triggered a two-hour data exposure
Meta		Agent misbehaviourMedium