Overview
- White House officials opposed Anthropic’s proposal to add about 70 more organizations to Mythos, which would lift total access to roughly 120, with Thursday’s reports citing security risks and worries about limited computing capacity.
- Senior aides are drafting guidance that could let select agencies regain use of Anthropic’s tools despite the Pentagon’s February supply‑chain risk label, as the national cyber office presses companies this week for concrete plans to counter AI‑driven attacks.
- Mythos is a Claude model built to spot and chain software flaws in major operating systems and browsers, so Anthropic is keeping it in a restricted preview called Project Glasswing for vetted tech and financial partners.
- Anthropic said it is investigating reports that a small group of unauthorized users in a private forum accessed Mythos, adding to concerns about controlling who can use the tool.
- Competitors and regulators are responding, with OpenAI starting a restricted rollout of its GPT‑5.5‑Cyber model and banks across Asia tightening checks as they prepare for AI that can find bugs faster than teams can patch them.