AgentPostmortem: A Public Registry of AI Agent Failures

Описание и применение

AgentPostmortem – это публичный реестр сбоев ИИ‑агентов, позволяющий документировать, искать и анализировать реальные инциденты, чтобы предотвратить повторения.

Задача

AI agents with write access cause real harm, but failures are undocumented and repeat.

Результат

AgentPostmortem creates a structured, moderated, and searchable record of AI agent failures, enabling teams to learn from past incidents and prevent repeats.

Рабочий процесс

• 1. Submit a failure case via the web form, providing agent, title, prompt, outcome, damage level, cost, tags, and optional evidence. 2. The submission is hashed for rate limiting, PII redacted, and stored with a pending status. 3. A moderator reviews the case for coherence, relevance, and abuse. 4. Upon approval, the case receives a sequential APM identifier and is published to the public feed. 5. Users browse, search, and vote on cases; data can be exported as CSV for further analysis.

Доказательства

AgentPostmortem is that record. It is a community-driven database of incidents where an AI agent caused real harm.

Источник

The stack is deliberately boring so the data outlives any hype cycle.

Источник