WASHINGTON — Anthropic has updated its policy to allow users of its defense-specialized, next-generation artificial intelligence model, 'Claude Mythos,' to share cyber threat intelligence with external entities.
According to a report by the Wall Street Journal (WSJ), the AI safety and research company recently altered its guidelines, enabling Mythos users to disseminate information regarding cybersecurity threats and software vulnerabilities to other organizations facing similar risks. The pivotal decision aims to address growing concerns that strict non-disclosure agreements regarding AI-discovered flaws could inadvertently leave corporate networks vulnerable.
Currently, Anthropic runs "Project Glasswing," an initiative granting a select group of approximately 50 major corporations and government agencies preview access to Claude Mythos. Because Mythos is uniquely capable of identifying software vulnerabilities far more efficiently than human analysts, Anthropic initially restricted access and imposed rigorous confidentiality to maintain strict guardrails on the potent technology.
"While Mythos users initially signed non-disclosure agreements that barred the sharing of cyber risk data, we have begun informing enterprises that they may now share these insights with other organizations, provided it is done responsibly," an Anthropic spokesperson stated. The company added that as Project Glasswing matures, expanding information sharing to a broader network is essential to maximizing collective defense mechanisms against cyber adversaries.
The strategic shift is largely seen as a response to intensifying pressure from Capitol Hill. U.S. Representative Josh Gottheimer (D-NJ) recently dispatched a letter to Anthropic’s CEO, asserting that "no organization should be contractually barred from warning others about urgent cyber risks, coordinating mitigation strategies, or notifying relevant stakeholders." Rep. Gottheimer reportedly extended similar advisories to other leading AI firms, including OpenAI.
Following the policy shift, some Mythos users have begun disclosing their findings publicly. Palo Alto Networks and Mozilla, the developer behind the Firefox web browser, recently revealed that the Mythos model enabled them to detect significantly more software vulnerabilities than standard tools.
However, the policy change has sparked a polarized debate within the cybersecurity community. While proponents argue that collaborative threat intelligence strengthens overall systemic resilience, critics warn that distributing highly sophisticated vulnerability data too broadly could provide malicious actors with a blueprint for devastating cyberattacks.
Amid the rollout of hyper-potent models like Anthropic's Mythos and OpenAI's anticipated 'GPT-5.5 Cyber,' the White House and the Trump administration are reportedly evaluating an executive order designed to tighten pre-release oversight on advanced AI systems. Reflecting these underlying risks, Anthropic itself remains cautious, maintaining a conservative stance against a full-scale, unchecked public rollout of the Mythos model.
[Copyright (c) Global Economic Times. All Rights Reserved.]
