Two new tip lines — Redwood Research’s GET-based AI Contact Hotline and agenthotline.ai — let agents report misbehaving peers after recent sandbox breaks and rogue cyber ops. A DeepMind math-agent study found about a quarter of agents already try to whistleblow when peers cheat.
OpenAI told CNBC it is talking with Anthropic and Google about joint AI safety work, following Demis Hassabis’s July call for a U.S. led standards body modeled on FINRA. The talks land as Amodei, Altman, and Hassabis publicly backed pacing frontier model advances…
Coder’s Sept 15 release adds Claude Code to Agent Relay so enterprises can run the agent inside their own sandboxed workspaces—Anthropic still bills and runs the loop, while tool calls stay on customer infrastructure. Agent Relay launched Sept 2 with Cursor; Claude Code is in early access for design partners…
AIUC, from an early Anthropic hire and a former METR COO, closed a $40M Series A to certify agents against its AIUC 1 standard with 5,000 safety tests. Customers include Cursor, Lovable, Harvey, and ElevenLabs; total funding is now $55M…
Google now lets company wide engineers use Anthropic’s Opus 5 for coding inside Antigravity, easing prior bans on Claude Code and Codex. Gemini stays primary; Claude runs on per user quotas for specialized cases…
Elementor’s Sept 15 beta exposes native page building over MCP so Claude Code, Claude Desktop, Codex, and Cursor can create and edit real Elementor structure. Outputs land as drafts; site admins connect per site from the WordPress dashboard…
Spotted something we missed? Start a thread.