Anthropic

Anthropic’s Claude Managed Agents offloads sandbox isolation, state persistence, and fault recovery to Anthropic, so developers only need to define agent logic. The three-layer decoupled architecture (Session/Harness/Sandbox) reduces p95 TTFT by over 90%, priced at $0.08/session-hour.

Anthropic launches Project Glasswing, investing $100M in AI credits and $400K in donations. Using the unreleased Claude Mythos Preview model, they discovered thousands of zero-day vulnerabilities in critical software worldwide, including a 27-year-old OpenBSD bug and a 16-year-old FFmpeg vulnerability.