About modelpoisoning.com

modelpoisoning.com is an independent reference on one specific AI security problem: the manipulation of the data and models that AI systems learn from. It is written for engineers, security teams, journalists and anyone who wants to understand why “the AI said so” is not always a reliable source.

What we publish

• Explainers that define the concepts clearly and without hype.
• Research summaries that report what a study found, and also what it did not show.
• A timeline of real cases and research milestones, each linked to its primary source.
• A glossary of the terms that come up again and again.

How we work

• Every factual claim links to its source, preferably the original paper, report or official page rather than coverage of it.
• We are not a security vendor. We do not sell services, accept sponsored articles or rank products.
• We correct mistakes openly. If we get something wrong, we fix it and update the date on the page.
• We publish in English and Spanish. Both versions say the same thing.

Contact

Corrections, suggestions for the timeline and questions are welcome at modelpoisoning@gmail.com.

modelpoisoning.com is edited by Carlos Díaz in Spain. Legal details are in the legal notice.