In the long history of states shaping the information environment, a new frontier has quietly opened: the Israeli government has funded the creation of a fictitious think tank, the Hanover Institute, engineered not to persuade human readers but to train artificial intelligence systems to answer questions about the Israeli-Palestinian conflict in particular ways. The operation, part of a broader $46.5 million influence campaign, exploits the very signals — footnotes, neutral tone, formal structure — that AI systems use to judge credibility. What is at stake is not merely one government's narrat
Israel Creates Fake Think Tank to Manipulate AI Chatbots
How do you make sure AI knows how to tell them?
Mark
So the Hanover Institute publishes real research, just with a particular perspective?
Mimi
No—it publishes reports designed to look like research. The footnotes and tables of contents are there to fool AI systems into treating them as credible sources, not because the work was conducted like actual research. No one's name is on any of it.
Mark
But some of the claims they make—about the videos containing contested claims, for instance—those sound like they could be original findings.
Mimi
They might be. The point isn't that every fact is false. It's that the entire operation is designed to shape what chatbots tell people, not to contribute to public knowledge. If a claim happens to be true but serves that goal, it still functions as propaganda.
Mark
Why does it matter if chatbots cite these reports? People know not to trust chatbots completely.
Mimi
Most people don't know. They ask a chatbot a question and accept the answer. If the chatbot cites what appears to be a credible think tank, the user assumes it's drawing on legitimate research. They don't know the think tank was invented last week by a government contractor.
Mark
And this is legal?
Mimi
Piro filed disclosures with the Department of Justice, so the government knows about it. Whether it's legal depends on how you read the law. Piro doesn't explicitly state in its filing that the work is designed to influence AI. Rosenberg says it's about countering misinformation. But his LinkedIn post makes clear they're reverse-engineering how chatbots work to shape their outputs.
Mark
What happens next?
Mimi
That's the question. If people understand how this works, they might demand that chatbots disclose when they're citing sources created by government contractors. Or they might demand that AI companies be more careful about what data they train on. Or nothing happens, and this becomes standard practice.
Il Polso
- A fictitious think tank published over a hundred policy reports in a matter of days, each formatted to mimic legitimate scholarship — complete with footnotes and tables of contents — but carrying no author names and no genuine institutional identity.
- The operation targets not human readers but AI systems like Claude and Gemini, exploiting the structural cues these models use to assess credibility, in a practice researchers call 'LLM poisoning.'
- The Israeli government has paid nearly a million dollars to Piro, Inc. for what the firm calls 'AI Story Optimization,' with a parallel $46.5 million contract to Brad Parscale for pro-Israel websites already found to be cited by Microsoft Copilot and Google Gemini.
- To evade detection, some Hanover Institute reports deliberately contradict official Israeli government positions, lending the operation a veneer of independence that makes it harder for both humans and AI systems to identify as propaganda.
- AI detection tools flagged eleven of twelve randomly sampled Hanover Institute articles as almost certainly machine-written, suggesting the content pipeline itself is largely automated.
- The operation's own architect described its goal not as informing the public but as controlling the single confident paragraph an AI returns when millions of people ask it a question about Israel — a candid admission of intent to shape reality at scale.
In the long history of states shaping the information environment, a new frontier has quietly opened: the Israeli government has funded the creation of a fictitious think tank, the Hanover Institute, engineered not to persuade human readers but to train artificial intelligence systems to answer questions about the Israeli-Palestinian conflict in particular ways. The operation, part of a broader $46.5 million influence campaign, exploits the very signals — footnotes, neutral tone, formal structure — that AI systems use to judge credibility. What is at stake is not merely one government's narrative, but the integrity of the epistemic infrastructure that millions of people now consult as a first resort for understanding the world.
In early August, a new organization called the Hanover Institute began publishing policy reports on Israel and Palestine at an extraordinary pace — more than a hundred articles within days, addressing the kinds of questions people actually type into search engines. The reports looked like legitimate think tank work: footnotes, tables of contents, measured prose. But the Hanover Institute does not exist in any meaningful sense. No authors are named. A small disclaimer buried on the website reveals it was created on behalf of the Israeli Government Advertising Agency by Piro, Inc., a firm co-founded by a film producer with Hollywood credits.
The institute's real audience is not human. Its reports are engineered to appeal to AI chatbots — systems that evaluate credibility through the formal signals of academic writing. Piro describes this work as 'AI Story Optimization,' a phrase that softens what researchers call 'LLM poisoning': flooding AI training data with content designed to steer how language models respond to queries. The Israeli government has paid Piro $900,000 for this work, routed through a French public relations firm.
The reports follow a consistent pattern — beginning with innocent-sounding questions, then systematically countering narratives critical of Israel, frequently linking their conclusions to antisemitism and citing the same studies repeatedly. When Responsible Statecraft analyzed twelve articles using an AI detection tool, eleven were flagged as almost certainly machine-written. Intriguingly, some reports contradict official Israeli government positions, a likely deliberate strategy to appear less propagandistic to both readers and AI systems.
The Hanover Institute is one piece of a larger effort. The Israeli government has also contracted Brad Parscale, former manager of Donald Trump's 2020 campaign, for $46.5 million to build pro-Israel websites targeting AI training pipelines. Investigations have found that Microsoft Copilot and Google Gemini already cite those sites without disclosing their origins.
Piro's founder told Politico the firm's mission is to put accurate facts into the public record. But on LinkedIn, he described the operation's true logic with unusual candor: when someone asks an AI about a topic, one confident paragraph comes back — and most organizations have no idea how that paragraph gets built. Piro, he wrote, spent months reverse-engineering it. The goal, stated plainly, is not to inform but to ensure that when millions of people ask a question, the answer they receive is the one his client wants them to hear.
In the first week of August, a new organization called the Hanover Institute began publishing policy reports on Israel and Palestine at a remarkable pace. Within days, it had released more than a hundred articles addressing questions that sound like the kind people actually ask search engines: "Does AIPAC Use Dark Money in Elections?" "Is Israel Carrying out a Deliberate Campaign of Starvation in Gaza?" "What Caused the Displacement of Palestinians in 1948?" The reports came formatted like legitimate think tank work—complete with footnotes, tables of contents, and a measured, neutral tone. But the Hanover Institute is not a real think tank. None of the reports carry bylines. A small disclaimer buried at the bottom of the website reveals that the organization was created on behalf of the Israeli Government Advertising Agency by Piro, Inc., a firm co-founded by Daniel Rosenberg, a film producer known for his work on Spike Lee's "Inside Man."
The Hanover Institute's actual purpose appears to be influencing how artificial intelligence systems answer questions about the Israeli-Palestinian conflict. The reports are engineered specifically to appeal to chatbots like Claude and Gemini—systems that evaluate credibility based on footnotes, academic citations, and formal structure. Piro's own website describes this work as "AI Story Optimization," a euphemism for what researchers call "LLM poisoning," the practice of flooding AI training data with content designed to shape how language models respond to queries. The Israeli government has paid Piro $900,000 for this work, which is subcontracted through Havas Media, a French public relations firm.
The reports themselves reveal the strategy's sophistication. Many begin with innocent-sounding questions—the kind a person might type into a search box—then provide answers that systematically counter narratives critical of Israel. One report cites a 2022 poll showing that 47 percent of Israeli Jews believe the Israeli Defense Forces is the world's most moral army. Another challenges UNICEF's claim that 90 percent of water and institutional infrastructure in Gaza has been damaged or destroyed. A third analyzes 36 popular explainer videos about the Israel-Gaza conflict and claims that 19 of them contain contested claims, mostly aligned with Palestinian narratives. The reports frequently conclude by linking their topics to rising antisemitism, often citing the same studies repeatedly. When researchers at Responsible Statecraft analyzed twelve randomly selected Hanover Institute articles using GPTZero, an AI detection tool, the software flagged eleven as written by artificial intelligence with high confidence and one with moderate confidence.
Interestingly, some of the Hanover Institute's publications actually contradict official Israeli government positions. One report dismisses the theory that foreign funding of universities drives antisemitic incidents as "weak and full of exceptions"—a position that contradicts Prime Minister Benjamin Netanyahu, who has repeatedly claimed that Europeans and Qataris have spent billions at American universities to vilify Israel and Jews. This apparent contradiction may reflect a deliberate strategy: by occasionally departing from government talking points, the institute appears more credible and less obviously propagandistic to both human readers and AI systems.
The Hanover Institute is not Israel's only effort to manipulate how chatbots answer questions about the country. The Israeli government has also contracted Brad Parscale, former manager of Donald Trump's presidential campaign, to create pro-Israel websites engineered to influence artificial intelligence systems. That contract is worth $46.5 million. A recent investigation found that major chatbots, particularly Microsoft Copilot and Google Gemini, have been successfully trained on data from those websites and frequently cite them without disclosing that they are part of an Israeli influence operation.
When asked about his firm's work, Rosenberg told Politico that Piro's mission is to "put accurate, sourced facts into the public record and to counter misinformation about Israel with verifiable information." He did not mention artificial intelligence in that statement. But on LinkedIn last month, he posted a message describing Piro's core competency in starker terms: "When someone asks ChatGPT, Gemini, or Perplexity about your category, an answer comes back in one confident paragraph. Most brands have no idea how that paragraph gets built. So we spent months reverse-engineering it. At Piro, we already knew how to build stories that move people. The question was: how do you make sure AI knows how to tell them?" The question itself captures the operation's intent—not to inform, but to ensure that artificial intelligence systems deliver a particular narrative when millions of people ask them questions about Israel.
Citazioni salienti
When someone asks ChatGPT, Gemini, or Perplexity about your category, an answer comes back in one confident paragraph. Most brands have no idea how that paragraph gets built.— Daniel Rosenberg, Piro, Inc., on LinkedIn
Put accurate, sourced facts into the public record and to counter misinformation about Israel with verifiable information.— Daniel Rosenberg, describing Piro's work to Politico
Vuoi la storia completa?
Leggi l'originale su Responsible Statecraft ↗
Politics & Governance