Israel’s Covert ‘Hanover Institute’ Operation Shows How States Are Now Gaming AI for Influence
Israel covertly funded a fake US think tank that pumped out more than half a million words of tailored content in nine days, optimized to be treated as authoritative by AI chatbots. The operation shows how governments are starting to weaponize the information diets of AI systems that millions already trust as neutral.
If you can quietly shape what an AI chatbot thinks is credible, you can bend the information diet of millions of people who rely on it. That is the logic behind a covert Israeli operation that built a fake US think tank designed not to sway human readers directly, but to feed the machines training on them.
Israel funded the creation of a bogus research institute branded as the “Hanover Institute,” which presented itself online as a US-based think tank, according to new disclosures. Over a span of just nine days, the site published 124 reports totaling more than 560,000 words—an output more akin to a data dump than a traditional policy shop. Every headline was crafted to mirror questions an AI user might type into a chatbot, such as queries about Middle Eastern policy or security dynamics.
Crucially, the site was built on a platform that advertises its ability to optimize content so that AI systems like ChatGPT, Claude and Perplexity will treat it as credible research. Rather than focusing on human subscribers or media citations, the operation appears to have targeted the training and retrieval pipelines of large language models, betting that if the content could pass as authoritative to algorithms, it would be surfaced—and repeated—to end users seeking answers.
The immediate human impact is subtle but real. Users turning to AI assistants for neutral explanations about complex conflicts may now be drawing on a knowledge base deliberately salted with state-backed narratives masquerading as independent analysis. Policymakers, students, journalists and ordinary citizens who lean on these systems for a first take on geopolitics may find their understanding quietly nudged in ways they cannot easily trace or question.
Strategically, the Hanover scheme lays bare a new front in information warfare: not just seeding social media or traditional media, but targeting the models that increasingly mediate what people see and believe. Governments have long funded friendly experts and think tanks to push their lines in Washington or Brussels; the difference here is scale, speed and opacity. An AI-optimized influence operation can flood the training pool with tailored content faster than a handful of real institutions could ever write—and with far less scrutiny.
The episode also poses hard questions for AI developers and regulators. If a state actor can fabricate an entire ecosystem of pseudo-scholarship in a week and get it ingested into widely used systems, then the line between open-source intelligence and weaponized noise becomes harder to draw. Security teams now have to think not only about guarding their models against hackers, but also about defending their training data against quiet capture by well-resourced propaganda campaigns.
This is a reminder that information integrity in the AI age is not just about fact-checking outputs; it is about who gets to define the corpus those outputs draw from. When a state can create a think tank out of thin air and teach machines to trust it, the traditional markers of credibility—location, branding, academic tone—become tools in an influence toolkit rather than guarantees of independence.
Key signals to watch now include whether AI providers move to detect and downgrade content from synthetic institutions like the Hanover Institute, whether other governments are exposed running similar operations, and what safeguards—technical or legal—emerge to protect training data from covert state manipulation. The next major influence campaign may never trend on social media; it may simply sit, quietly, in the memory of the systems people now treat as impartial guides.
Sources
- OSINT