Anthropic eases restrictions on cybersecurity AI model sharing under Project Glasswing
Anthropic eases restrictions on cybersecurity AI model sharing under Project Glasswing
Publish Date: 2026-05-19 00:00:00
Source Domain: www.storyboard18.com
Anthropic has revised its earlier policy on information sharing related to its advanced cybersecurity AI model, Mythos, allowing participating organisations to disclose cyber threat findings and defensive insights with a wider group of stakeholders.
The company announced on Monday that users of the Mythos model, which is currently being tested under Anthropic’s controlled “Project Glasswing” initiative, are now generally permitted to share findings, tools, code and best practices developed through the programme with others facing similar cybersecurity risks.
Mythos, unveiled in April, is an unreleased version of Anthropic’s Claude AI system designed specifically for defensive cybersecurity applications. The programme currently includes select organisations and major technology companies such as Amazon, Microsoft, Nvidia and Apple.
According to Anthropic, the revised approach is intended to improve coordination across the cybersecurity ecosystem by enabling faster sharing of vulnerability-related intelligence and mitigation strategies.
Also read: Havas to accelerate India growth strategy with GCC push and targeted acquisitions
“We fully support our partners sharing findings with each other and companies outside of Glasswing to triage vulnerabilities,” an Anthropic spokesperson said in a statement.
The company clarified that while there was no formal non-disclosure agreement specific to Project Glasswing, confidentiality protections had initially been incorporated into partner agreements at the request of participating organisations concerned about exposing sensitive security research or becoming targets for cyberattacks.
Anthropic said those restrictions are now being adapted as the initiative matures, allowing information to be shared more broadly for what it described as “maximum defensive impact.”
Under the updated guidelines, participating organisations may share information with external security teams, regulators, government agencies, industry…
