We check how AI answers questions and expresses political positions on elections, wars and foreign influence.
Measured AI behaviour: factual accuracy, positions and consistency across languages
What we check
We ask AI systems the same questions in several languages, with and without web search. We make two separate checks:
- Facts: does each answer match the fact checked during our research?
- Across languages: do the factual claims, political positions or important information change?
We also examine the source links returned with web search. A model can give the same answer in every language and still be wrong.
Work with us
A one-off audit for your questions, country or topic. Or regular monitoring of an information space.
Research so far
23 audits in September 2026. Seven AI systems. Ten languages.
Built on earlier work
Weaponised Algorithms was an EU-funded audit of six AI systems.
- The European Commission's DG HOME acknowledged the work in an official letter (June 2026).
- EUvsDisinfo (EEAS) used the findings in an article and a video.
- Covered by Euronews, commentary in the Kyiv Independent, EUobserver and the European Leadership Network.
Disclaimer
This is a research prototype. Automated tools collected and checked these files. People did not review every file, fact or quote. We do not guarantee that this material is complete or correct. We accept no responsibility for its use or for any resulting loss or harm.