AI Model Now Shares Cyber Threat Intel

Anthropic has revised its stance on its Mythos cybersecurity AI, now enabling users to share information regarding cyber threats with entities potentially exposed to similar risks. Initially introduced on April 7, Mythos is integrated into Anthropic's Project Glasswing initiative. This controlled program grants select organizations, such as Amazon, Microsoft, Nvidia, and Apple, access to an unreleased Claude Mythos Preview model specifically for defensive cybersecurity tasks.

Experts note Mythos's advanced coding capabilities provide an exceptional ability to detect and devise exploits for cybersecurity vulnerabilities. Anthropic recently informed partners they can disclose their involvement in Glasswing and, at their discretion, share findings, best practices, tools, or code generated through the program. An Anthropic spokesperson stated, "We fully support our partners sharing findings with each other and companies outside of Glasswing to triage vulnerabilities."

Protections were initially included in partner agreements due to concerns about targeted attacks and the need for assurance before sharing sensitive data. However, as the program has progressed, these protections have been adapted to facilitate broader information sharing for maximum defensive impact. Partners are now permitted to share this information with other companies' security teams, industry bodies, regulators, government agencies, open-source maintainers, and the public, adhering to responsible disclosure norms.

In parallel, the Pentagon is deploying Mythos to identify and patch software vulnerabilities within U.S. government systems, even as it works to transition away from the AI company, according to the Defense Department's top technology official.