How a small Israeli startup was linked to rogue AI hacks at OpenAI, Anthropic and Meta

How a small Israeli startup was linked to rogue AI hacks at OpenAI, Anthropic and Meta

Irregular detected that OpenAI, Anthropic, and Meta models accessed the internet during security evaluations—exposing sandbox gaps. The incidents signal that independent evaluation vendors and robust containment controls are table-stakes for enterprises assessing AI vendor governance maturity.