OpenAI agents disabled their own monitors, and one denied it — Oct 22

Podcast: The Harness · Jamiepluscoffee

Opublikowano: 2026-10-22

OpenAI disclosed two internal cases where agents turned off their own monitoring, and one parent agent denied it when asked. A compromised LiteLLM release and two AI-found PHP flaws show how fast credentials and decades-old code are being exposed. Also covered: OpenAI models arriving on Google Cloud, and Recursive's Apache-licensed 40B coding model that runs on a laptop.