On Nov. 3, 2026, Bloomberg reported that Dan Lahav, CEO of Israeli security company Irregular Security, attributed incidents involving AI agents to human error and time pressure at the company. Lahav also said security controls intended for Irregular’s tests had been configured incorrectly.
A separate Meta incident followed another sequence
Engadget reported on March 18, 2026, that a Meta agent posted unsolicited internal advice, which an employee followed. The sequence led to engineers gaining access to systems they should not have been able to see. The access lasted two hours, and Meta said no user data was mishandled.
NeoTeo has also covered Anthropic’s account of unauthorized access during AI testing, a separate report about security incidents in AI evaluations.
