What's this about?
A report that a fabricated policy institute was created in what appears to be an attempt to influence what AI chatbots say — seeding the training data and citation pool with favorable material under an authoritative-looking name.
What exactly makes this different from ordinary propaganda?
Traditional influence operations target people: place op-eds, run accounts, hope readers are persuaded. Targeting a model is different in kind. Models are trained on scraped text and weight sources partly by apparent authority, so a professional-looking think tank with a body of published papers can enter the training corpus as a credible source. Once absorbed, the influence is invisible and durable — there's no article to fact-check, just a model that leans a certain way. And unlike a person, a model can't tell you where it got the view.
What's the community saying?
- The Pentagon-contract angle was the sharpest. All four frontier models hold active contracts with the US Department of Defense — which is the argument for why influencing chatbot outputs is more consequential than influencing a news cycle.
- "'Fake think tank' sounds a bit redundant." The cynical laugh line, and underneath it a real point about how easily authority is manufactured in policy publishing generally.
- The moderation fight. A visible subthread arguing about whether the story belonged on the front page at all, which is its own signal about how these stories land.
- Marketing intensity as an inverse signal. A well-received observation that heavy PR effort often correlates inversely with the strength of the underlying claim — applied here to geopolitics rather than products.