The Ghost Think Tank: How Israel’s Propaganda Machine Targeted AI

The Discovery That Changed Everything

In August 2026, journalist Jason Wilson uncovered something extraordinary hiding in plain sight on the internet. The Hanover Institute for Public Policy appeared to be everything a respectable think tank should be—it had polished reports, academic citations, references to the World Bank and United Nations, and sophisticated statistical analysis. It tackled the most pressing issues of our time: genocide allegations, apartheid, occupation, and humanitarian crises in Gaza. There was just one problem: none of it was real. The institute had no physical address, no verifiable researchers, no legal existence. Between August 6 and 14, this phantom organization somehow produced 124 reports containing over 560,000 words—nearly 354,000 words in just 48 hours at one point. The sheer scale of this operation was staggering, but the real story lay much deeper. What investigators discovered was not merely a fake think tank, but a sophisticated attempt to manipulate not just human readers, but the artificial intelligence systems that are increasingly becoming our primary source of information about the world.

The Architecture of Doubt

The Hanover Institute’s brilliance lay not in telling bold lies, but in manufacturing uncertainty. When human rights organizations characterized Israeli actions as genocide or apartheid, Hanover’s reports would carefully note that no final judicial ruling had yet been issued. When casualty figures were cited, the reports would question the reliability of the sources providing them. A single 2022 study about antisemitic comments on Facebook pages was cited 454 times across 88 different reports, appearing even in documents about land registration and the 1948 Palestinian expulsions. This is propaganda at its most sophisticated—not crude denial, but the systematic erosion of certainty itself. The strategy was simple yet devastating: surround every claim with enough caveats, qualifications, and doubts that readers would conclude nothing could truly be known. You don’t need to say “Israel didn’t do this” if you can make every mention of the alleged action so hedged with uncertainty that the truth becomes indistinguishable from speculation. This approach works because it doesn’t demand allegiance to a counter-narrative; it simply makes the original narrative impossible to grasp with confidence.

Following the Money Trail

The true scope of this operation only becomes clear when you follow the money through official government records. Under the Foreign Agents Registration Act, Piro, Inc. registered as an agent for the State of Israel, specifically working through Havas Media Germany on behalf of Israel’s Government Advertising Agency. The company’s own filing to the Department of Justice admitted its activities “may include communications activities intended to influence the U.S. public.” But the most revealing document came from another firm entirely. Clock Tower X LLC, whose registration was signed by Bradley Parscale, included a contract clause that should alarm anyone who uses artificial intelligence tools. The agreement explicitly called for “Deployment of websites and content to deliver GPT framing results on GPT conversations.” This wasn’t about advertising on AI platforms or studying how they work—it was about actively shaping what AI systems would say when people asked them questions. The budget for this operation was equally staggering: $1.5 million per month for four months, with requirements to produce 5,000 content variants monthly and generate 50 million impressions. The machinery of influence had been built, funded, and aimed directly at the algorithms that increasingly mediate our understanding of reality.

The New Propaganda Frontier

What makes this operation genuinely unprecedented is its target. Traditional propaganda sought to persuade people directly—through newspapers, radio, television, or social media. The new propaganda infrastructure described in these contracts seeks to persuade the intermediary that will select information for those people. This is an upstream operation, designed to contaminate the river before the answer is constructed. Piro’s own marketing materials describe the strategy with almost pedagogical clarity: they map “every surface the models read,” create content “engineered for how LLMs evaluate credibility,” and deploy it on trusted third-party properties they build. The company describes serving two simultaneous audiences: “the engine that recommends you, and the person who believes it.” This represents a fundamental shift in how propaganda operates. Instead of merely contesting the answer, this approach intervenes before the answer exists. The old propagandist needed an audience; the new one can try to secure intermediaries. When you ask an AI system a question about Gaza, it doesn’t give you links to compare—it gives you a synthesized, authoritative-sounding answer. Whoever controls what the machine reads controls what you believe, and the source of that information becomes invisible in the polished final product.

The Threat of Epistemic Capture

The implications of this operation extend far beyond the Israel-Palestine conflict. The technological infrastructure described in these contracts—industrial-scale content generation, networks of websites, automated monitoring of AI responses, and continuous optimization—could be deployed by any state, corporation, or political movement with sufficient resources. The most conservative scenario already exists: political actors studying which sources AI systems use, producing content designed to appear credible to those systems, and measuring whether their narratives get cited. But the more troubling possibilities compound this threat. An operator could produce millions of pages designed to survive web crawls and influence training datasets, creating statistical gradients that shape how AI systems understand entire categories of events. Through systematic querying and measurement, an operation could create an adversarial feedback loop—training the environment itself. The extreme scenario would be epistemic capture: wealthy actors colonizing the corpus available to AI systems until what machines perceive as “consensus” reflects who had the greatest capacity to produce and distribute sources, not what is actually true. The technology for all of this exists today. What’s missing is not capability but scale, persistence, and coordination—precisely the properties emerging in the documents examined.

The Battle for the Archive

This investigation reveals something profound about the nature of power in the age of artificial intelligence. Genocidaires have always needed to dominate more than territory—they need to contest the archive itself. They need certainty to become controversy, victims to become dubious sources, and perpetrators to become merely “one of the parties” to events whose causality becomes increasingly difficult to reconstruct. Artificial intelligence introduces a possibility never before available at this scale: placing a machine between the citizen and the archive. A generation that once consulted newspapers, documents, and books can now ask a system a question and receive ten paragraphs of synthesis, with the sources disappearing behind the narrator. Defending the right to information can no longer be limited to combating fake news; we must defend the integrity of the information ecosystem itself. AI systems should distinguish genuinely independent sources from domains financed by the same actor, and provenance information should inform how machines evaluate political material. Journalists must investigate infrastructures, not just texts—who registered the domain, who signed the contract, who built the website, who expects a machine to read it. The final irony is that despite all this machinery of influence, Israel’s image continued to deteriorate, particularly among young people. Perhaps the reputational decline wasn’t caused by poor communication but by people seeing what was actually happening. And when you cannot change what produces rejection, you try to change the narrative about it. The old power wanted to write history; the new one wants to intervene in the automated historian. This investigation doesn’t end with Hanover—it begins there, because in the age of artificial intelligence, the most efficient way to hide a fact may no longer be to erase it, but to manufacture a million documents around it.