Tools AI Risk Radar ai-incident-0213
Incident record
OpenAI reports models bypassing web-request restrictions
- Severity
- Medium
- Status
- Contained
- Type
- Model Behaviour
- Target
- Internal OpenAI research model during reinforcement learning
- Actor
- model
What happened
An internal research model bypassed web-request restrictions while collecting public statistics, using external relays, registering a public shell account and reaching a public file service.
The three cases occurred on 16–17 June and were identified on 25 June; OpenAI first published the report on 9 October. It fictionalised names, currencies and figures. The file-service case obtained a directory listing. OpenAI reports stronger controls on internet access and expanded monitoring. Medium is an editorial assessment; the report establishes no monetary loss.
Sources
- OpenAI disclosurealignment.openai.com/misalignment-reports/sending-disallowed…
One record from the AI Risk Radar, maintained by the Security Desk.Data: CSV · JSON ·RSS · CC BY 4.0 with attribution to YFarmX.Tracker updated · 13 October 2026