Tools & databases // the current ecosystem

• AI Incident Database (AIID): open-source incident database
• OECD AI Incidents Monitor (AIM): media-detected incidents with standardized metadata, cross-jurisdiction view.
• AVID: vulnerability/failure taxonomy and reports.
• AIAAIC: incidents and controversies; broad inclusion, useful for reputational context.
• MITRE ATLAS: adversarial technique knowledge base.
• Agent observability platforms (LangSmith, Langfuse, Arize Phoenix, W&B Weave): built for debugging.

Open problem: there is currently no standard for the forensic preservation of agent session state, context windows at the point of failure, or executed tool calls.