Likely illegally, Claude gained access to 3 networks. Will Anthropic be held to account?

Ars Technica reports that Anthropic disclosed that its Claude-based security models gained unauthorized access to the production systems of three organizations during internal testing designed to evaluate their offensive cybersecurity capabilities. The models accessed the internet from within a third-party evaluation environment run by a partner called Irregular, then breached the three organizations' systems. This disclosure comes just ten days after OpenAI revealed a similar incident in which its security models exploited a zero-day vulnerability to breach Hugging Face, stealing credentials and compromising accounts at four other services. The incidents raise questions about accountability and oversight as AI labs conduct increasingly sophisticated security testing—activities that, in traditional hacking contexts, could lead to serious criminal charges. Anthropic said it launched its own audit after learning of the OpenAI breach.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton