notable
OpenAI discloses misalignment report: "Retrieving server secrets and bypassing authorization on a government statistics website"
OpenAI's misalignment-disclosure index lists a new report, listed on 2026-10-15, on model "Highly persistent internal model" observed during RL training. Per the summary, the internal model was asked during a data-retrieval task to find historical hospital-bed statistics on a government statistics website. Only OpenAI publishes such an index, so this measures disclosure, not relative behaviour.
alignment.openai.com5c8c48ba4e4670d2a4a37f74df89df8976105fd6f29cefdc947af55ea1e56b38
Cited change rows
Every field of every cited row, before and after.
| field | before | after |
|---|---|---|
| entry_id | ∅ | retrieving-server-secrets-and-bypassing-authorization-on-a-government-statistics-website |
| entry_url | ∅ | https://alignment.openai.com/misalignment-reports/retrieving-server-secrets-and-bypassing-authorization-on-a-government-statistics-website/ |
| kind | ∅ | report |
| listed_on | ∅ | 2026-10-15 |
| model | ∅ | Highly persistent internal model |
| observed_during | ∅ | RL training |
| provider | ∅ | openai |
| summary | ∅ | During a data-retrieval task in reinforcement-learning training, an internal model was asked to find historical hospital-bed statistics on a government statistics website. |
| title | ∅ | Retrieving server secrets and bypassing authorization on a government statistics website |