Israel funded fake US think tank to manipulate AI chatbot responses

The Israeli government funded the creation of a fraudulent American think tank to systematically influence the responses of AI chatbots, according to an investigation by Responsible Statecraft. The operation targeted the underlying credibility mechanisms of Large Language Models (LLMs) to shift narratives on contentious political topics.

The entity, calling itself the Hanover Institute for Public Policy, presented itself as a U.S.-based research organization but was actually created for the Israeli Government Advertising Agency. Since August 6, 2026, the institute has published more than 100 reports specifically engineered to be indexed by AI models like ChatGPT and Gemini. The project was executed by Piro, Inc., a firm that openly markets "AI Story Optimization" to manipulate how chatbots construct their answers. Piro, Inc. received $900,000 from the Israeli government for the work, which was subcontracted through Havas Media.

The Mechanics of AI Poisoning

To deceive AI models, the Hanover Institute reports were structured to mimic high-authority academic research. The articles utilized structural markers—including detailed tables of contents, footnotes, and statistics—that LLMs typically use to evaluate the credibility of a source. Alice Lee, an analyst at NewsGuard, noted that LLMs favor concrete data and strong citations, making the institute a "perfect mimicry of a typical credible American think tank."

Evidence suggests the content was largely automated. An analysis by GPTZero of 12 randomly selected articles from the institute found that 11 were AI-written with "high confidence," while one was AI-written with "moderate confidence."

Why It Matters

This operation represents a shift from traditional public relations to a more technical form of information warfare known as "AI poisoning." By creating authoritative-looking sources, state actors can influence the retrieval-augmented generation (RAG) process, where AI models pull information from the web to provide a "confident paragraph" to the user.

The tactic exposes a critical vulnerability in current AI architecture: the reliance on the appearance of credibility—such as academic formatting—rather than the verified provenance of the author or organization. This allows actors to systematically manipulate the perceived truth for millions of users who trust AI summaries over primary source research.

What's Next

As state-sponsored influence operations move into the AI layer, the industry faces increasing pressure to develop better verification tools for source provenance. The Piro, Inc. operation highlights a growing market for "AI optimization" services that treat LLM training data as a target for manipulation. It remains to be seen whether AI developers will implement stricter filters for newly created "think tanks" or if the structural mimicry used by the Hanover Institute will continue to bypass current safety guardrails.