SocietyAI glossary
Data poisoning
Deliberately inserting into training data elements designed to disrupt the model that will learn from it: making it associate a trigger word with a hidden behaviour, or degrading its results. Research has shown that a few hundred booby-trapped documents can be enough, even in a gigantic corpus. Some artists also use it to protect their work from being harvested.
In other words
Slipping a few fake pages into a student's textbook: they learn them along with the rest, and the mistake resurfaces on exam day.
Alyriaz articles that mention it
- The week AI went looking for its raw material
Weekly round-up for 3–9 August 2026: data takes centre stage, prices keep tumbling, and opacity gains ground on the regulatory front. We connect the dots.
- What is data poisoning? When a few hundred texts are enough to trap a model.
Data poisoning involves inserting elements into a corpus that are designed to disrupt a model trained on it. We explain the mechanism, why it works with very small quantities, and the debate it raises.
- AI labs are buying books by the pallet, and destroying them to scan them.
AI labs are bulk-buying physical books published before 2022, touted as the last source of guaranteed human-written text. The industrial scanning destroys the volumes. We explain why this rush is happening, what it says about the state of data, and what it threatens.