New from the Lab·The Compass — an open moral reasoning standard for AI, tested across frontier modelsExplore →
Production AI Institute · PSF v1.1 open standard
AI Right-To-KnowAI Data Use IndexCheck My AI ToolsPolicy Change WatchAgent ReadinessPublic BenchmarkContactGlobal standard · Worldwide
HomeResearchContribute
Research Contribution

Submit a production AI incident

PAI collects anonymised incident data contributed voluntarily by practitioners. Your incident data helps the community understand what goes wrong in production AI, informs PSF domain evolution, and is included in aggregate in our published incident pattern analysis.

What we ask for

System descriptionType of AI system, model provider (can be anonymised), deployment context (internal/external, industry, approximate scale)
Incident timelineWhen was the incident first detected, when was it resolved, and how long was the system in the affected state
Root causeYour assessment of the primary failure mode, mapped to PSF domains if possible
Detection methodHow was the incident detected — monitoring alert, user report, manual review, or other
ResolutionWhat was done to resolve the incident and prevent recurrence

Anonymisation policy

All identifying information — organisation name, system name, model version, and any personnel details — is removed before the data is used. All contributions are reviewed before inclusion in published analysis. You will be contacted before any case data is used in a published analysis.

Submit via email

Submit your incident report via our contact formwith the subject line “Incident contribution”. You can use free-form text; we will structure the data during review. If you prefer a structured template, ask for one in the form and we will send it to you.

Contributors to published analyses are acknowledged by name or pseudonym at their choice. If you contributed to an incident that led to a significant change in PSF domain design, we will note this in the relevant framework note.

Related: Incident Patterns analysis · Production Safety Framework