Anthropic has launched a restricted preview of its most advanced AI model, “Mythos,” limiting access to approximately 150 organizations involved in its Project Glasswing cybersecurity programme. Rather than a general public release, the company is controlling the distribution of this powerful tool to prioritize defensive security and infrastructure hardening.
Understanding Claude Mythos
Mythos is designed as an advanced model capable of autonomously identifying software vulnerabilities, generating exploit code, and assisting security teams in patching critical flaws. Because the model can identify thousands of bugs across browsers and operating systems, Anthropic has classified it as a high-risk system with significant offensive cyber capabilities.
“The core issue is that a model that can help defenders can also accelerate attackers if it spreads too widely,” according to industry analysis regarding Anthropic’s safety strategy.
Controlled Access for Critical Infrastructure
Anthropic has framed the access as an invitation-only, defensive effort. By keeping the technology restricted, the company aims to test the model within controlled settings to harden critical infrastructure before considering any broader deployment.
Fast Facts: Project Glasswing and Mythos
- Access Scope: The program includes roughly 150 organizations, comprising large technology and infrastructure firms. PDF
- Participating Entities: Access has been granted to major players including Apple, Google, Microsoft, AWS, and NVIDIA. PDF
- Responsible Disclosure: Partners are permitted to share findings under strict responsible-disclosure rules to ensure the model is used for security purposes.
A New Pattern in AI Safety
Anthropic’s strategy reflects a shift in how AI developers manage powerful systems. By treating cyber capability itself as the primary risk, the company is using a “restricted-access” model to ensure that developers use these tools to strengthen defenses rather than facilitate unauthorized exploits.
Social Call-to-Action: As AI technologies become increasingly powerful, what measures should global firms take to ensure these tools are used ethically for national security? Let us know your thoughts on our official NTA social media channels.
