Anthropic has made the unprecedented decision to withhold its most advanced AI model, Claude Mythos Preview, from public release, citing its cybersecurity capabilities as being too powerful and dangerous if made widely available. The model, announced on Tuesday as part of a new industry initiative called Project Glasswing, has demonstrated an ability to autonomously identify tens of thousands of previously unknown zero-day vulnerabilities, including critical flaws in every major operating system and web browser.
The scale of discovery is staggering. Claude Mythos Preview found a 27-year-old vulnerability in OpenBSD, one of the world's most security-hardened operating systems, and a 16-year-old bug in the ubiquitous FFmpeg media library that had been missed despite automated tools testing the relevant code line five million times. It also identified weaknesses in major cryptography libraries like TLS, AES-GCM, and SSH, and a myriad of web application vulnerabilities.
Due to the dual-use risk—where the same capabilities that secure systems could be weaponized for attacks—Anthropic is restricting access. Instead of a public API, the company has assembled a coalition of 12 launch partners, including Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorgan Chase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks. An additional 40 organizations that build or maintain critical software infrastructure will also receive access. Anthropic is committing $100 million in usage credits and $4 million in direct donations to open-source security organizations to support the defensive effort.
The model represents a significant leap in autonomy. According to Anthropic, Mythos identified nearly all disclosed vulnerabilities and developed working exploits for many entirely autonomously, without human steering. In one demonstration, it chained several Linux kernel vulnerabilities to escalate from user access to full machine control. On benchmarks, Mythos scored 83.1% on the CyberGym vulnerability reproduction test, compared to its predecessor Claude Opus 4.6's 66.6%. On the SWE-bench Verified software engineering test, it scored 93.9% versus 80.8%.
"AI capabilities have crossed a threshold that fundamentally changes the urgency required to protect critical infrastructure from cyber threats, and there is no going back," said Anthony Grieco, SVP and Chief Security and Trust Officer at Cisco. "The old ways of hardening systems are no longer sufficient."
The announcement comes amidst heightened industry warnings. OpenAI, in an industrial policy paper released just before Anthropic's announcement, identified AI-enabled cyberattacks as one of the two most immediate near-term risks from advanced AI. Sam Altman warned that soon-to-be-released models could enable a world-shaking cyberattack within the year. Real-world incidents already show the threat: a Chinese state-sponsored group previously exploited Claude's capabilities to infiltrate roughly 30 organizations, and a hacker used Claude in attacks against Mexican government agencies in February.
Project Glasswing is a strategic bet that giving defenders a head start will create a durable advantage. "The window between a vulnerability being discovered and being exploited by an adversary has collapsed — what once took months now happens in minutes with AI," said Elia Zaitsev, CTO of CrowdStrike.
The initiative is not without complications. Anthropic's own security was recently tested when it accidentally exposed nearly 2,000 source code files last month, and the existence of Mythos was first revealed through a leak from an unsecured data cache. The company is also navigating a dispute with the U.S. Department of Defense over military use of Claude.
Anthropic has committed to publishing a public report on lessons learned within 90 days. However, the race is on. "No one organisation can solve these cybersecurity problems alone," Anthropic stated. "The work of defending the world’s cyber infrastructure might take years; frontier AI capabilities are likely to advance substantially over just the next few months." Project Glasswing represents Anthropic's opening bid that AI-powered defense can outpace AI-powered attack.
