124 reports in nine days: Inside Israel’s fake think tank bid to shape ChatGPT

The Guardian has uncovered an elaborate scheme by an Israeli-backed think tank to shape the responses of AI systems like ChatGPT. The Hanover Institute for Public Policy published a staggering 124 reports in just nine days, averaging 4,500 words per report. These reports, which appeared between August 6 and 14, focused on contentious allegations involving Israel, such as torture of Palestinian prisoners, war crimes, and genocide.

The institute, which has no legal existence or identifiable researchers, published its reports without bylines and with vague claims of legal jurisdiction. Almost every report took the form of a question a user might input into an AI chatbot. The reports often framed evidence in a way that downplayed criticism of Israel, citing courts' lack of final verdict on allegations and attributing figures to "a party to the events."

The site's technical infrastructure suggested it was designed with AI systems in mind, carrying an llms.txt file that helped AI understand the site's content. Cleveland-Stout, a researcher who investigated the site, noted that despite its academic appearance, the material frequently concluded that discussion of criticism could fuel antisemitism.

The strategy behind this work could have a lasting impact, as the material could be absorbed and potentially cited by AI systems without proper fact-checking. While the Hanover Institute's reports were not part of Common Crawl's July index, other websites operated by Clock Tower X, run by former Trump campaign manager Brad Parscale, were.