Published: April 10, 2026
Anthropic has announced Claude Mythos Preview, the most capable model the company has ever built. Rather than making it broadly available, Anthropic is restricting access to defensive cybersecurity use cases through a new initiative called Project Glasswing. The model has already been used to discover thousands of zero-day vulnerabilities across major operating systems and web browsers.
What is Claude Mythos Preview
Claude Mythos Preview is a general-purpose frontier model that Anthropic positions as a significant step above its current Opus 4.6 tier. The model achieves 93.9% on SWE-bench Verified, 77.8% on SWE-bench Pro, 82% on Terminal-Bench 2.0, and 97.6% on USAMO 2026. Each of these scores represents a double-digit improvement over both Opus 4.6 and GPT-5.4.
What sets Mythos apart is its performance on security-specific tasks. The model scores 83.1% on CyberGym, the leading cybersecurity benchmark, compared to 66.6% for Opus 4.6. It can also convert 72.4% of identified vulnerabilities into working exploits, a capability that previous Claude models consistently failed at.
Project Glasswing and security discoveries
Anthropic launched Project Glasswing alongside the Mythos announcement. The initiative allows partner organizations to use Mythos Preview specifically for defensive security scanning. Participating companies include Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, Microsoft, and Nvidia.
In the weeks leading up to the announcement, Anthropic used Mythos Preview internally to identify thousands of zero-day vulnerabilities across every major operating system and every major web browser. Several of these vulnerabilities had existed undetected for years. The oldest was a 27-year-old bug in OpenBSD. Anthropic has committed up to $100 million in usage credits to the initiative and donated $4 million to open-source security organizations.
Why Anthropic is not releasing it publicly
Anthropic has stated that it does not plan to make Claude Mythos Preview generally available. The company’s reasoning centers on the model’s dual-use potential; the same capabilities that make it effective at finding and patching vulnerabilities could also be used to discover and exploit them. Anthropic wants to develop new safeguards before deploying Mythos-class models at scale.
This approach marks a departure from the typical AI release cycle, where companies compete to make their latest models available as quickly as possible. Anthropic is instead prioritizing controlled deployment, accepting the competitive disadvantage in exchange for what it frames as responsible stewardship.
What this means for the AI landscape
The Mythos announcement signals that at least one major AI lab has reached a capability level where unrestricted public access raises genuine security concerns. It also validates the argument that frontier AI capabilities will increasingly intersect with national security and critical infrastructure protection.
For enterprise users, Project Glasswing offers a new tool for proactive vulnerability detection. For the broader AI industry, it raises questions about how future frontier models will be distributed and who will have access to the most capable systems.
For more information, see Anthropic’s official announcement at anthropic.com/glasswing.
