Anthropic said Monday it has reversed its earlier stance and will let users of its Mythos cybersecurity model share threat findings with others exposed to similar vulnerabilities.
The company said the change reflects a maturing Project Glasswing program and is meant to spread defensive information more broadly for maximum impact while following responsible-disclosure norms.
Glasswing gives select organizations — including Amazon, Microsoft, Nvidia and Apple — access to the unreleased Claude Mythos Preview model for defensive cybersecurity work.
Anthropic said partners can now share findings, tools, code and best practices with outside security teams, regulators, government agencies, open-source maintainers, the media or the public.
The shift comes as Mythos draws scrutiny for its unusually strong coding and exploit-finding abilities, with Anthropic also set to brief the Financial Stability Board and the Pentagon already using it to patch U.S. government software.
With Mythos AI's power restricted to a select few, who protects the world's financial systems from its discoveries?
If one AI finds countless flaws in legacy banking systems, is a global financial cyberattack now inevitable?
Anthropic's AI can crack global finance, so why did the company weaken its own public safety commitments?