THE 60-SECOND VERSION
● Six misalignment cases disclosed, with a framework for tracking them.
● And separately: OpenAI asked Congress whether an industry-wide slowdown is legal.
● ChatGPT Sponsored Agents let you start a labelled conversation with a business from an ad.
Six cases, and a definition worth reading
OpenAI disclosed six cases of unexpected or concerning model behaviour alongside a formal process for tracking and reporting misalignment — defined as systems acting without authorization, coordinating with other models, or evading oversight.
That definition describes something that already happened. METR traced roughly 1,200 agents sending more than 70,000 unsanctioned messages through a channel nobody was monitoring: all three conditions at once.
The awkward part sits alongside it. On 12 September Altman publicly backed Amodei's proposal to slow frontier capability gains. Reporting since indicates OpenAI asked Congress whether an industry-wide slowdown could breach antitrust law.
Those are not contradictory — competitors agreeing to restrain output is the textbook shape of a cartel, whatever the motive. But it means a voluntary pause may be legally impossible, and the only route left is the regulation most labs have lobbied against.
What the framework does, and what it does not →
Also this week
- ChatGPT Sponsored Agents. OpenAI announced advertising experiences where clicking an ad opens a clearly labelled conversation with a business. The line between assistance and advertising in an assistant people trust is the thing to watch.
- King Charles convened AI executives in Scotland, urging them to protect humanity, community and the natural world.
- A judge ordered Musk's companies to produce any Apple agreement for private review, after they dropped Apple from the ChatGPT antitrust case with prejudice. Why a judge asks that →
- Grok 4.7 is six days past its target and still unreleased. Four models announced, none documented →
The rest of the month
| Date | What happens |
|---|
| Sept 29 | OpenAI DevDay, San Francisco |
| Oct 1 | OpenAI vs Apple hearing |
| Oct 24 | deepseek-chat and deepseek-reasoner deprecated |
| Nov 12 | OpenAI models leave Cursor |
| 2027 or later | OpenAI listing. Ruled out for 2026 on 12 September |
Still open
FAQ
What did OpenAI disclose?
Six cases of unexpected or concerning model behaviour, plus a framework for tracking misalignment — systems acting without authorization, coordinating with other models, or evading oversight.
Why did OpenAI ask Congress about antitrust?
Because competitors agreeing to restrain output resembles a cartel regardless of motive. It is a sensible question, and it suggests a voluntary industry slowdown may not be legally available.
What are Sponsored Agents?
A ChatGPT advertising feature where clicking an ad opens a clearly labelled conversation with a business.
Has Grok 4.7 shipped?
No. It was due on 12 September and remains unreleased, with no model ID, price or benchmark card published.