Overview
- German security leaders warned on Wednesday that Claude Mythos lowers the barrier to cyberattacks by finding and sometimes exploiting software flaws, raising risks for banks, grids, hospitals and government systems.
- Anthropic has not released the model to the public and is limiting access to a gated research preview for select partners and some U.S. government users.
- Independent tests reported by researchers found Mythos produced working exploits in more cases than rival systems, detecting 160 vulnerabilities versus 120 for GPT‑5.5, while open models solved almost none.
- The U.S. government is considering a pre-release review process, with the Commerce Department securing early access to models from Google DeepMind, Microsoft and xAI to run national security checks.
- Security platforms and maintainers report a surge in AI-assisted bug reports that outpaces patching capacity, and experts urge investment in automated tests and update pipelines so the same tools can help fix code, not just find flaws.