AI for Cyber Defenses? Anthropic's Mythos Sparks Cautious Debate

Anthropic says Claude Mythos can autonomously detect and exploit software vulnerabilities at scale, prompting the company to withhold public release over cybersecurity misuse fears; a controlled preview (Project Glasswing) is being offered to select firms to test defensive use, while experts note the tool's potential to accelerate vulnerability discovery and patching but warn about cost, safety, and the risk of abuse in the hands of attackers before defenses catch up.