Overview
- Anthropic remains barred from Pentagon contracts after a federal appeals court in Washington, D.C. refused to pause the Defense Department’s “supply chain risk” designation, which directs defense vendors to prove they do not use the company’s Claude models.
- Responding with a defensive push, Anthropic introduced Project Glasswing, giving vetted partners access to the unreleased Claude Mythos Preview model and $100 million in credits to test and harden their own systems.
- The company said early runs of Mythos found thousands of serious software flaws, including a 27‑year‑old OpenBSD bug and a 16‑year FFmpeg issue, and it pledged a public report within 90 days while coordinating with U.S. agencies.
- To scale capacity and improve reliability after surging demand, Anthropic hired longtime Microsoft leader Eric Boyd to run infrastructure and is pursuing large U.S. data‑center builds.
- Peers are moving in the same direction, with OpenAI reported to restrict a high‑security model to select partners and Meta rolling out the smaller, efficient Muse Spark model as tech leaders call out compute bottlenecks and plan massive AI spending.