For journalists

AI Misalignment Watch tracks AI misalignment incidents and collects every report about each one, from the labs, institutes and research groups that ran the tests. Use it to find the original sources fast, then link to them.

What you can use

• Incident tracker. Each incident states the model, when it happened, when it was first reported, who was affected and what happened, with every report pinned to it, starting with the report from the company that made the model. See the incidents.
• Plain-language summaries of each report, always linked to the original. Incident and Report Summaries.
• Charts you can download as images or embed, free to use with credit. Analytics.
• Open data for your own analysis: reports (CSV), reports (JSON), incidents (CSV), incidents (JSON).
• Feeds. RSS and Bluesky.

How to cite

Please link to the original report first. To credit this site, write:

AI Misalignment Watch, misalignmentwatch.com, accessed Oct 16, 2026.

What we do and do not do

• We summarise original reports. We do not investigate, and we do not add claims or predictions to the summaries.
• Every report is from the organization that ran or studied the model. We do not use news or social media.
• A person approves every incident and every summary before it goes live. Weekly Summaries label analysis and opinion separately from facts.
• If we get something wrong, the original wins. We correct mistakes quickly.

Contact

Questions, corrections or a request for a chart: corrections@misalignmentwatch.com. Logos: mark (PNG), share image (PNG).

The site has tracked reports since Dec 1, 2021.