Anthropic Allows Partners to Share Cybersecurity Findings, Raises Security Concerns

Anthropic, an AI startup, has introduced new measures that could potentially heighten risks associated with AI usage and security.

On April 7, Anthropic announced its “Mythos” cybersecurity model as part of “Project Glasswing,” a controlled initiative allowing select organizations like Amazon, Microsoft, Nvidia, and Apple to use the unreleased Claude Mythos Preview model for defensive cybersecurity purposes.

Experts believe that Mythos’ coding capabilities could potentially identify cybersecurity vulnerabilities and devise ways to exploit them.

Recently, Anthropic began informing partners about their involvement in Glasswing, permitting them at discretion to share findings, best practices, tools or code developed through the program.

An Anthropic spokesperson stated that they fully support partners sharing findings with each other and companies outside of Glasswing to triage vulnerabilities. Confidentiality protections were included in agreements signed by participating companies after partners sought assurances before sharing sensitive information and expressed concerns about being targeted by attackers.

As the program has matured, these protections have been adapted to ensure key information can be shared broadly—including outside the program—for maximum defensive impact. Mythos is now being deployed by the Pentagon to find and patch software vulnerabilities across the U.S. government as it transitions away from Anthropic.