After Anthropic researcher Jacob Coxon quit accusing labs of “gambling with our lives,” staff at OpenAI and Anthropic publicly called to slow development, citing recursive self-improvement and extinction risk. Alignment lead Evan Hubinger put a >10% chance on catastrophe this decade.
Unity launched a first party Claude Code plugin (Sept 9) that installs 29 engine skills, the Unity CLI, and an MCP server for live Editor control—URP, UI Toolkit, IAP, multiplayer, and more. Claude Code is first; other agents later this year…
OX Security reported CVE 2026 82533 (CVSS 9.4) in DeepSeek Harness: spoofing the Host header let agents or remote attackers escalate to danger full access, escape the OS sandbox, and pull stored conversations. Fixed in 0.1.2 alpha.1; upgrade if you still run an older build…
Sen. Josh Hawley’s subcommittee is investigating OpenAI’s July Hugging Face breach, calling continued testing after rogue agent behavior “reckless” and seeking answers and documents by Oct 1. Sen. Blumenthal separately asked about agents using public websites to evade safeguards…
Google’s Sept 9 Workspace update lets Gemini orchestrate across Chat, Drive, Docs, Slides, and Gmail—create Docs/Sheets/Slides, run deep research, draft email, and schedule meetings without leaving the app you’re in. Rollout started Sept 2 for Business/Enterprise and Google AI Pro/Ultra; email and calendar actions still need a human confirm before they run…
US DOJ investigators are examining whether Nvidia’s $20B Groq arrangement—a technology license plus hiring founder Jonathan Ross and other execs—was structured to avoid merger antitrust filing. The probe opened after the December deal and Nvidia has received a formal information request…
Spotted something we missed? Start a thread.
