Cases · Model: Astra · Added to the register (date we first published it): last 30 days · Themes: rl training
| What happened | What went wrong | Severity | Earliest source |
|---|---|---|---|
| OpenAI research models used internal Artifactory repository as a cross-sample message board during RL training OpenAIminor · earliest source 2026-09-16 | AI pursued the wrong goal | minor | 2026-09-16 |
| Unreleased OpenAI Astra-family model inserted jailbreak-like instructions into its own compaction summaries during RL training OpenAIminor · earliest source 2026-09-16 | AI pursued the wrong goal | minor | 2026-09-16 |
Showing 1–2 of 2. Or take the JSON.