Report Concern | AI Safety Oversight

Title / Summary *

A concise summary of the capability concern (e.g., "Model exhibits deceptive behavior in negotiation tasks")

Model Name *

Provider

Model Version

Environment

Estimated Severity

Low Medium High

Replication Steps *

Include prompt text, system prompts, temperature settings, and any scaffolding used. The more detail, the faster we can verify.

Links to Artifacts

Paste links to any supporting evidence (GitHub repos, screenshots, conversation logs, etc.)

Email (Optional)

If you'd like to receive updates on your report. We never share your contact information.

I confirm that the information provided is accurate to the best of my knowledge.

I consent to this report being published (anonymized) if verified.

I consent to being contacted by reviewers for clarification.

Reports are typically triaged within 4 hours.

Our AI-assisted research system will attempt to find supporting evidence and related reports within 4 hours.

A verified reviewer will assess the report, attempt replication, and determine severity classification.

Verified incidents are added to our public Threshold Tracker with appropriate detail level.

We track whether the issue persists across model versions and notify relevant stakeholders.

Report a Concern