Investigation Reveals Israeli-Funded Fake Think Tank Manipulating AI Systems To Shape Global Perceptions of Gaza

A new joint investigation has exposed what its authors describe as a “criminal disinformation enclave” — a sophisticated Israeli-funded machinery that produced a fake think tank, hundreds of think tank front pages and hundreds of thousands of fabricated words in a matter of days — all designed not only to mislead human readers but to manipulate the artificial intelligence systems that will increasingly select and synthesize what humanity knows. The Hanover Institute for Public Policy appeared on the internet with every attribute of a respectable policy-research center, from academic citations to references of World Bank and Amnesty International, dealing precisely with the most sensitive questions surrounding Israel’s international reputation: genocide, apartheid, occupation, hunger in Gaza, Palestinian displacement, war crimes, antisemism, and the political power of AIPAC. But The Guardian’s investigation, published August 26, 2026, by Jason Wilson, revealed that behind the façade stood no verifiable legal entity, no physical address, no identified researchers, and a production capacity that should have been impossible: 124 reports in nine days, totaling more than 560,000 words, with 73 documents — nearly 354,000 words — produced in just 48 hours on August 12–13. Of those 124 reports, 123 were built around questions phrased extraordinarily similarly to the kind a person might type into Google or ask an artificial intelligence directly, leaving an uncomfortable paradox: whether the work was entirely automated or was produced by a hybrid army of humans, the same sobering conclusion stands — this was an industrial operation.

The money trail, tracked through documents filed under the Foreign Agents Registration Act (FARA) with the U.S. Department of Justice, reveals who built it and who paid for it. Pira, Inc., a public-relations firm, registered its relationship with the Israeli government in explicit detail: it identified the Israel Government Advertising Agency (LaPam) as the represented party, and stated that its services — “strategic communications and media relations services” — may include “communications activities intended to influence the U.S. public.” The firm, which markets a service called “AI Story Optimization,” acknowledged that its industrial purpose is to map “every surface the models read” — Reddit conversations, YouTube transcripts, forums, comparison sites — and rank those sources according to how much influence they hold over what AI engines say. Piro says it creates content specifically engineered to meet “how LLMs evaluate credibility.” But the chilling breadth of the documented scheme comes from the parallel player, Clock Tower X LLC (FARA registration No. 7649). Its documents name Havas Media as its prime contract holder “Supervised: State of Israel,” and signatory of the contract is Bradley Parscale, under penalty of perjury, dated September 18, 2025.

Even more remarkable than the existence of the contracts is what the contracts actually require. The Initial Statement of Work that governs Clock Tower LLC’s work specifies a billable base contribution of $6 million from the client for only four months, with an additional $1.5 million per month contributed by the client during that period. The theater of operations required at least 100 core creative assets per month, 5,000 monthly content variants (at least 80% of it tailor-made for TikTok, Instagram, YouTube and podcasts aimed at Generation Z), and it required paid campaigns to deliver at least 50 million impressions per month. That section alone was enough to reveal the industrial scale of the operation. But then the contract goes one step further — the one that genuinely changes the field. In a section dedicated to “Search and Language Operations,” the annex requires the contractor to run monthly SEO campaigns using an AI-assisted search platform to improve “visibility and ranking of relevant [political] narratives.” And it explicitly commands: “Deployment of websites and content to deliver GPT framing results on GPT conversations.” That is not speculation nor academic theorizing. It is a contract term. The report does not claim.
that Clock Tower had control of ChatGPT, or that OpenAI collaborated with the campaign. But it demonstrates that the intent to deliberately influence the framing of artificial-intelligence conversations was a described goal, and it was budgeted for.

The strategic revolution this work represents is subtler and profoundly more dangerous than the classic matter of stuffing the internet with lies. The authors explain that old propaganda violence invented False facts; the new form — well demonstrated by Hanover institutional report — instead uses limited truth as its core tactics: when human-rights organizations called Israel’s actions genocidal or apartheid, the outlet injected the caveat that no final judicial ruling had yet been issued; when casualty data was referenced, it added doubt about who had presented the numbers; and the favorite academic study about online antisemism was cited 454 times in 88 different reports, even in contexts about land registration or hunger in Gaza, where it was irrelevant. This is what is known in the academically as framing — a precise operation capable of determining what is in the foreground and background of any argument, where doubt is injected, and what causal relationships are even perceptible. They explain it through a corporate Piri slogan: there are two audiences for such an effort — “the engine that recommends you, and the person who believes it.” The new infrastructure first tries to persuade the intermediary (AI systems) that will tell people the news. And rather than simply controlling facts, they attempt to colonize the very information space that AI models statistically weight when building their answers. If AI models learn from a noisy web environment packed with millions of words, patterns of “doubt” and of Palestinian delegitimization can become a “statistical edge” deeply encoded — even if a literal lie never appears in any one document.

The investigation raises the most challenging question about what it calls “provenance laundering”: when an entity finances twenty web domains that seem independent but are, in fact, vehicles for one single source of money, and when machines count every domain as one, twenty voices appear, one behind the numbers. The possibility of what the author calls “epistemic capture” — a form of information domination in which material control of the income distribution translates into what the machine will later represent as “consensus” or “reality in the mind of the model.” The danger is heightened by a concerning detail: Common Crawl, the massive web archive that was over 2 billion pages, and commonly used as raw material of datasets feeding large language models, included pages belonging to sites linked to Clock Tower during its July 2026 crawl. The investigation itself carefully warned that establishing the presence of the Clock Tower sites in Common Crawl does not establish that the LLM was trained on them — and it explicitly says of the methodology, “the precise data corresponds to the directory, not the source of the transformation.” Yet it creates a clear argument that the technological infrastructure is in place for both an “adversarial feedback loop” and for attempts to train the environment AI breathes.

This clandestine machinery comes amid the continuing legal and historical reckoning over Israel’s conduct in Gaza. In January 2024, the International Court of Justice ruled that “at least some of the rights” claimed by South Africa — those of Palestinians in Gaza to be protected from acts of genocide — were “plausible” and ordered provisional measures bind them; a May 2024 order reaffirs this determination. The campaign is therefore not a “learning operation” of a brand. It is, the investigation concludes, “aimed at contesting the memory.” The authors are clear not to attribute the legal status of genocide to the ICJ findings; they instead draw from this series straight forward: “Haplessly impossible also even contest memory — They need certainty, twenty years from now, to be able to appear controversial again; the unexpected.” In what they call the new, era of the struggle over the archive, the final reported sentence captures the change of the battlefield: under traditional propaganda you used to write history; under the new one, you participate in the automatic historian. “If Hanover were an autonomous and successful single person, it was original.” But the real verdict lies beyond the particulars of Israel’s strategy: when one state masters the industrial production of facts for the machines, it can be followed by another, or a corporate, or an ideologue of any kind, and the box of intelligence put inside the very definition of a true fact. Ordinary users have no ability to check who got there first. The report warns in its closing lines: “It is no longer enough to defend truth against censorship… Traditional propaganda needed to reach millions of people. The new propaganda can try to reach the machine and let it do the rest.”