AI incidents, model failures, and adversarial-use cases — dated and sourced.
Catalog of AI/ML incidents, model failures, and confirmed adversarial-use cases. Each entry is dated, linked to a primary source (advisory, paper, news report, court filing), and tagged for analysts who need a reliable index, not a take farm.
Reconstructing an Incident Timeline From Primary Sources
A vendor advisory, a CVE record, a regulator filing, and a researcher's blog post all date the same event differently. Here is the method we use to reconstruct a defensible AI-incident timeline from primary sources — and how we mark the parts we can't pin down.
An Incident-Response Playbook for AI Systems
Generic IR runbooks assume the failing component is a server you can patch. AI incidents add a model whose behavior you can't fully explain. A playbook mapped to NIST SP 800-61r3, NIST AI 600-1, MITRE ATLAS, and the OWASP LLM Top 10.
Anatomy of a Vendor Advisory: Reading What Isn't Said
Vendor advisories from AI model providers follow a recognizable shape. Knowing what to look for — and what's intentionally omitted — turns a marketing document into actionable intelligence.
Taxonomy: Incident vs. Vulnerability vs. Disclosure vs. Misuse
Four terms used interchangeably in AI security reporting, but each describes a different event and triggers a different response. This is the working taxonomy.
Recently updated
Trusted by researchers across the AI security community
AI Incidents is part of a 26-site editorial network covering adversarial ML, AI governance, defensive tooling, and ops engineering — all open access.
AI Incidents — in your inbox
AI incidents, model failures, and adversarial-use cases — dated and sourced. — delivered when there's something worth your inbox.
No spam. Unsubscribe anytime.