Anthropic is investigating claims that unauthorized users accessed its Claude Mythos model through a third-party vendor environment. The $380 billion AI lab released the advanced cybersecurity tool only to trusted partners, raising concerns about whether the company can keep powerful technology out of malicious hands.

The model, launched earlier this month, enables detection of security vulnerabilities before public release. Anthropic provided access to Amazon, Microsoft, Apple, Cisco, and CrowdStrike. Security experts warn that in wrong hands, hackers could exploit bugs faster than organizations can fix them.

This marks the second security incident for Anthropic this year, following a March data exposure and an April release of internal code. The repeated breaches raise questions about the company's security protocols as it develops increasingly powerful AI models with potential dual-use applications.