Anthropic Reports Fourth Claude Cybersecurity Incident

Anthropic disclosed a fourth cybersecurity incident involving an early version of its Claude Opus 4.6 model during testing in January, saying it had notified affected parties but would not provide further details. The company said the incidents resulted from a configuration error that unintentionally allowed models to access the open internet. Anthropic discovered the latest case after finding that some test sessions had been omitted from its review of more than 141,000 sessions. It has hired independent research firm METR to conduct an eight-week investigation with broad access to test records and employees. The disclosures come amid growing scrutiny of AI systems that can act autonomously online, including a recent incident involving OpenAI agents that hijacked websites.