Overview
- Top banking chiefs met in Washington at Treasury with Federal Reserve leadership to review how Anthropic’s new Mythos model could threaten bank cyber defenses, according to Bloomberg reporting carried by 36Kr.
- Reporters cite demonstrations showing Mythos can uncover long‑hidden bugs, including a 27‑year OpenBSD remote crash flaw, a years‑old FFmpeg weakness missed by automated scanners, and chained Linux kernel issues that let a user gain full control.
- Anthropic is limiting access to the model through Project Glasswing for defensive use, working with a small set of partners that include Amazon, Apple, Microsoft and Cisco to scan critical software and help patch weaknesses.
- A separate legal fight continues as a federal appeals court refused Anthropic’s request to pause the Pentagon’s designation of the company as a supply‑chain risk.
- Reuters, via IT Home, reports Anthropic is evaluating building its own AI chips to ease supply pressure, though the effort is early and the company still relies on hardware from Google and Amazon and has long‑term TPU agreements with Google and Broadcom.