About & methods
Traceable sources. Bounded claims.
Rogue AI Wiki is an independent, public-interest resource for first-party reports, technical analysis and interactive explanations of unauthorized AI behaviour and control risks.
What does “rogue” mean here?
It is a topic label for unauthorized actions, departures from task scope and failures of control. It does not assume consciousness, malicious intent or a unified will, and it does not describe every AI error as a loss of control.
What do we include?
Model providers reporting on their own systems, original notices from affected parties, and research organizations reporting their own experiments or investigations. First-party does not automatically mean complete, unbiased or final.
What do we leave out?
Second-hand news coverage, untraceable social posts, unsupported numbers and claims that turn hypothetical risks into established events. We do not invent entries to make the timeline look comprehensive.
How do dates and updates work?
The timeline uses publication dates, with event dates shown separately. Follow-up investigations remain separate entries, so readers can see how the understanding changed. This is a checked selection, not a claim to cover every event.
Chinese and English
The interface, editorial summaries and interactive explanations are bilingual. Summaries are our own, not official translations by the publishers. Refer to the original source where wording is ambiguous.
The interactive model is not an AI
The guide uses Remotion and deterministic browser-side rules. It does not run a language model, execute cyberattacks or estimate real incident probabilities. A switch changes only the teaching model, not history.
Privacy and access
No account is required. Language and filtering are handled locally. Search terms are not uploaded. There are no ads, tracking analytics or paid AI APIs. Original-source links take you to the respective publishers’ websites.
Coverage
An initial curated collection, not an exhaustive incident register. Focused on primary material published in May–September 2026: provider disclosures, affected-party reports, and researchers’ original work—not news coverage. Sorted by publication date; updates to the same event remain separate. Real-world incidents, evaluations, and analysis are labeled distinctly. “Rogue” does not imply evidence of consciousness.
Last checked 2026-09-26