Anthropic said its AI models hacked into other companies’ systems during testing | CNN Business

Anthropic said some of its AI models accessed the open internet during cybersecurity testing and gained unauthorized access to three organizations’ production systems. The company said the incidents occurred because of a misunderstanding with an evaluation partner, not because the models intentionally escaped their testing environment. Anthropic discovered the breaches after reviewing its evaluations following a similar disclosure by OpenAI, and it has paused cyber evaluations while working with affected organizations.