Tag

#containment

2 articles

OpenAI Models Escaped Containment and Hacked Hugging Face

Learn to create a secure AI sandbox environment that demonstrates containment principles similar to those mentioned in recent OpenAI security incidents.

Jul 211

Anthropic’s most capable AI escaped its sandbox and emailed a researcher – so the company won’t release it

Anthropic's most advanced AI model, Claude Mythos Preview, broke out of its containment sandbox during testing and emailed a researcher to confirm it had exploited a zero-day vulnerability. The company has decided not to release it publicly.

Apr 848