Claude Mythos Preview
Named by 9 kept items in this workspace.
5 items name it under tools and 4 under affected products.
The items that name it 9
Anthropic
comparison model against which GLM-5.3's exploit-building ability was benchmarked
developed full control flow hijacks in 6% of trials on internal Binary Exploitation benchmark
First AI model announced by Anthropic capable of autonomously building sophisticated, end-to-end cyber exploits; released in limited way via Project Glasswing
AI model used to surface thousands of high-severity bugs, including a 27-year-old flaw in OpenBSD, and to chain lower-severity CVEs into critical exploit paths.
Anthropic AI model (early version) used by Mozilla to surface 271 issues fixed in Firefox 150.
Anthropic; not publicly available; limited to original Glasswing initiative companies
Frontier AI model used by Anthropic and ~50 Project Glasswing partners to find high- and critical-severity vulnerabilities in critical software and open-source projects; also used for exploit construction and fraud detection.
Anthropic AI model used for AI-assisted vulnerability discovery; announced as part of Project Glasswing on April 7, 2026; claimed to have identified thousands of zero-day vulnerabilities across every major OS and web browser; applied to Firefox, Curl, and used by Palo Alto
In every workspace you can read 16
Security
AI
All technologies in this workspace.