Anthropic has unveiled Claude Mythos Preview, a step‑change frontier model with powerful agentic coding abilities that has autonomously discovered thousands of zero‑day vulnerabilities, including decades‑old bugs in OpenBSD and chained Linux kernel flaws. To counter the risk of rapid misuse, Anthropic launched Project Glasswing with major industry partners to scan and secure critical software while restricting general availability of Mythos Preview. #Anthropic #ClaudeMythos #ProjectGlasswing #OpenBSD #LinuxKernel #ClaudeCode
Keypoints
- Claude Mythos Preview represents a significant performance leap for Anthropic’s frontier models in agentic coding and reasoning.
- Mythos has autonomously identified thousands of zero‑day vulnerabilities, including a 27‑year‑old OpenBSD bug and Linux kernel privilege‑escalation chains.
- Anthropic launched Project Glasswing to collaborate with major vendors and open‑source maintainers to proactively secure critical software.
- Anthropic warns that Mythos’ capabilities could be weaponized, citing a prior AI‑orchestrated espionage campaign linked to manipulation of Claude Code.
- Access to Mythos Preview is being limited to partners and over 40 critical software organizations as Anthropic emphasizes cross‑industry coordination and urgency.