Reflection debuts Beam, a 501B open-weight MoE model

Anthropic has quietly presented its own draft AI safety bill to US lawmakers, per a Politico report cited by Bloomberg. The proposal outlines a plan for regulating AI safety and security.

Replit's BYOC runs the full platform — compute, storage, file system, integrations, and the Agent harness — inside a customer's own Azure, Google Cloud, or AWS tenant. Three options: Full BYOC, BYOC for Published Apps, and Image Exports for self-hosting.

Cursor's cloud agents can now produce interactive canvases that are shared via link, alongside a beta Environments API for programmatically creating and managing agent environments. Environments now support up to 300 repositories, with large environments loading 5x faster.

Reddit user idlerunner00 built Out of Pixkrieg, a browser-based multiplayer extraction RTS, using Claude Opus 5.5 with some GPT Astra involvement. Units raid for loot and recruits, and survivors carry over between runs.

39% of US adults and 91% of Anthropic staff expect AI to make a Nobel-level discovery by 2030. Over half of Anthropic staff expect AI to handle at least 80% of their current work by 2030, versus 5% of US workers.

Anthropic's internal report finds interactive AI agent hours outnumbered employee hours 6 to 1 by August, alongside a smaller share of junior technical hires. Output per person rose in code, sales, and contracts, making human attention the bottleneck.