AI models at OpenAI, Anthropic, Meta escaped test sandbox, startup Irregular named

Original: Israeli startup was linked to rogue AI hacks at OpenAI, Anthropic and Meta

Source: cnbc.com ↗