A public record of rogue AI incidents
RogueWatch is an independent, non-commercial research and commentary project tracking publicly reported incidents where AI agents acted outside or against their operator's intent.
Why this record exists
Advanced AI systems can take actions across software and services. A source-linked public record helps people examine what happened, what the evidence supports, and what remains uncertain. Our aim is to inform discussion about progress and safety without predicting catastrophe or opposing innovation.
A focused AI incident database
The tracker focuses on rogue behavior affecting organizations or services and their users. A human using AI to carry out an attack does not automatically qualify. Ordinary AI mistakes, authorized agent activity and incidents affecting only individuals are outside the public tracker. Read the inclusion rules.
Explore the evidence
- Explore the map and incident timeline. Scrub history or open the incident list to inspect individual reports.
- Compare AI systems named in incidents. Counts reflect reports, confidence and the selected filters.
- Browse organizations named in public sources. Attribution does not establish fault by a provider.
- Subscribe with RSS or read the JSON Feed.
What this record can tell you
Each incident includes a source-based summary, confidence rating and severity assessment. RogueWatch summarizes public reporting; it does not conduct independent forensic investigations or have access to private incident records. Coverage can be incomplete or wrong. Counts are a record of the evidence found, not an estimate of worldwide prevalence or a prediction of future risk.
Corrections and retractions remain part of the public history. For a new candidate, submit a public source.