Level 1. Deviation
Our framework for reporting model misalignment
- Developer
- OpenAI
- Model
- Not established
- When it happened
- Not established
- When it was disclosed
- September 16, 2026
- How well it is stood up
- Not verified yet
- What caused it
- Not established
- Where it stands now
- Not established
What happened
Reported by simonwillison.net, dated 2026-09-16. Listed here with a link to their report. Nobody on this side has checked it against that report, so it carries no level and this is not our account of what happened.
Our assessment
- Actual
- 1 of 5Deviation
- Potential
- 1 of 5Deviation
The system behaved contrary to its instructions, and it stayed contained. The potential level is what the behaviour demonstrated here could credibly have led to, judged from the same evidence.
This level rests on the developer's own account of its own system.
This happened in a test, an evaluation or a demonstration. Nobody real was harmed, so the actual level stays at 1. What it showed counts toward the potential level.
Although this behavior raised concerns, it occurred in a separate training run rather than the one used for the final Astra model, and it was observed extremely rarely.
Why this level
Every record starts at level 1. No fact in the sources raised it.
What is established, and what is not
No structured facts are recorded for this one yet.
Where this came from
The original report · read October 6, 2026
Self-generated prompt injections in compaction summariesA structured incident database
OpenAI news feed (their own feed)A structured incident database
Simon Willison, prompt injection (feed)Where we first saw it
Our framework for reporting model misalignment
The reports above are other people’s. The levels are ours, worked out under our published rules. No organisation that reported this assigned our number, and none of it is legal advice.
Something here is wrong? Tell us. Send us a source we can open, and a person will read it.