Descriptions:
Anthropic has announced Claude Mythos Preview, an unreleased frontier model formerly codenamed Capiara, alongside Project Glass Wing — a critical software security initiative joined by Amazon, Apple, Broadcom, Cisco, Google, JP Morgan Chase, the Linux Foundation, Microsoft, Nvidia, and Palo Alto Networks. The company has explicitly declined to release Mythos publicly, citing the model’s demonstrated ability to autonomously find and exploit software vulnerabilities at a level surpassing all but the most elite human security researchers.
On benchmarks, Mythos achieves 93.9% on SWE-bench Verified, outperforming Claude Opus 4.6, Gemini 3.1 Pro, and GPT-5.4 by significant margins. More concretely, it is the first model to autonomously complete a corporate network attack simulation that experts estimated would require over 10 hours of human effort. In internal red-teaming, Anthropic used Mythos to identify thousands of zero-day vulnerabilities across every major operating system and browser — including a 27-year-old flaw in OpenBSD (widely regarded as one of the most hardened operating systems in existence) and a 16-year-old vulnerability in ffmpeg that automated testing tools had executed against five million times without catching.
YouTuber Wes Roth walks through the system card and contextualizes why Anthropic chose disclosure over deployment, quoting Cisco’s independent assessment that AI capabilities have “crossed a threshold that fundamentally changes the urgency required to protect critical infrastructure” — a conclusion Cisco reached after evaluating Mythos directly as part of the Glass Wing coalition.
📺 Source: Wes Roth · Published April 08, 2026
🏷️ Format: News Analysis
