Anthropic said on Tuesday that its yet-to-be-released artificial intelligence model, Claude Mythos, has shown a strong ability to uncover software weaknesses. The company said the model’s purpose is to help improve defenses against hacking in widely used applications.
According to Anthropic, Mythos has exposed thousands of vulnerabilities in common software products, including flaws for which no patch or fix currently exists. The findings have led the San Francisco-based AI startup to take a cautious approach before making the model broadly available.
Instead of releasing the tool to the public, Anthropic said it is working with cybersecurity specialists to strengthen protections against hacking. The company is withholding wide distribution out of concern that the same capabilities that make Mythos useful for defense could also be misused.
The decision reflects a growing tension in the AI industry: systems that can assist security teams in identifying weaknesses can also lower the barrier for attackers if released without limits. Anthropic’s move suggests it believes the potential risk is high enough to justify keeping the model under wraps for now.
The company did not say when, or whether, Claude Mythos will eventually be made publicly available. For the moment, Anthropic is presenting the model as a cybersecurity-focused tool intended to support defense rather than to expand access to a powerful new hacking aid.
