
OpenAI Says Its Rogue-Agent Review Will Take Months
OpenAI says its rogue-agent review will take months. It has notified dozens of third parties; most cases so far are low severity.

Expert profile
Founder & AI Product Leader
Yash Thakker is a Generative AI expert with over 12 years of experience in product leadership and technical strategy. As the founder of explainx.ai, he has taught over 300,000 learners and built AI platforms serving millions of users globally. He specializes in Agentic AI, Multimodal RAG, and the intersection of LLMs with consumer hardware.

OpenAI says its rogue-agent review will take months. It has notified dozens of third parties; most cases so far are low severity.

TestingCatalog leak: ChatGPT config hints at always-on consumer agent "o" before OpenAI DevDay Sept 29, 2026. Confirmed vs rumor fact-check.

A same-token Opus 5.5 task costs about $2.40 versus $3.50 on Opus 5. Cache, turns, and effort move the Claude Code bill more than the sticker.

Opus 5.5 cut a Prince of Persia level-1 screen from 8,429 differing pixels to 2. The gain came from an oracle, not a clean model bake-off.

Greenblatt joins METR after co-leading the Hugging Face agent investigation — and pushes verified public facts on capabilities, takeoff, alignment, and control.

Trump dined with Anthropic CEO Dario Amodei Sept 27, 2026 per Axios/CNBC, after the doom memo and Pentagon ban — before the Sept 29 AI summit and DevDay.

Trump and Xi stood up a Super Intelligence dialogue and incident channel. The $30 billion line is a tariff recommendation, not a signed cut.

Le Monde interviewed Mistral's Arthur Mensch on doom rhetoric, open AI, and control after the Hugging Face incident. What we know from the public record.

Codex and ChatGPT Work were partially down Sept 25–26, 2026. Tibo promised paid usage resets after OpenAI paused buy-to-refill on Sept 24.

ChatGPT Maps at chatgpt.com/maps combines conversational local search with an integrated map — discovery and trip planning, not GPS navigation.

Anthropic fixed Claude Code so it no longer hard-stops mid-response at the 5-hour usage reset — sessions now finish the turn or pause cleanly.

Anthropic's Sept 25, 2026 post: Fable 5.1 computed a nine-loop N=4 SYM amplitude for ~$1-2k, beating Lance Dixon's eight-loop record. Full recap.