Israel's ad agency, through Havas and Piro, stood up the…

Filed

September 12, 2026, 07:00 UTC

Actor

Israel Government Advertising Agency (LaPam), via Havas Media Germany GmbH and Piro, Inc. (FARA 7732), publishing as the Hanover Institute for Public Policy

Target

U.S. users of ChatGPT, Perplexity and other retrieval chatbots, plus search indices that treat Hanover reports as independent research on Gaza, the IDF, antisemitism and anti-Zionism

Source

Unidentified Source blurred: contractors, cutouts, buried disclosure

Confidence

documented Primary documents, court records, official disclosures

Techniques

Corpus poisoning Synthetic media Narrative laundering Disclosure theater Transfer Manufactured consensus

Where

United States, Israel, Gaza

Channels

hanoverinstitute.com unsigned data reports
llms.txt and question-formatted headlines aimed at LLM retrieval
Piro AI Story Optimization product copy
FARA 7732 informational-materials filings through 4 September 2026
Havas Media Germany subcontract chain from LaPam
ChatGPT and Perplexity citation surfaces
Footer disclosure that does not travel inside model answers

Summary

Piro, Inc., FARA registration 7732, opened 2 June 2026 to represent Havas Media Germany GmbH on behalf of the Israel Government Advertising Agency (LaPam). Exhibit AB records a 30 April 2026 digital-storytelling work order. Politico Influence on 14 August tied Hanover Institute reports to a $100,000 informational-materials campaign; 404 Media reported a $900,000 April invoice plus $100,000 in June. The Guardian counted 124 unsigned reports and more than 560,000 words published 6-14 August, most as chatbot-style questions. The live site still carries the FARA footer. Informational-materials filings continued through 4 September 2026. Politico's tests found ChatGPT and Perplexity citing Hanover on Gaza, anti-Zionism and antisemitism. Daniel Rosenberg said the job was putting sourced facts in the public record. Israel's foreign ministry told the Guardian it does not run influence operations in the United States. This brief scores the method, not the underlying Gaza claims.

Analysis

The load-bearing object this cycle is not another Claude factory already in this vault. It is a live FARA-registered content mill that dresses as a U.S. think tank so retrieval systems, not only voters, will cite Israel's preferred answers.

Primary record is Piro, Inc., FARA registration 7732, opened 2 June 2026. The foreign principal is Havas Media Germany GmbH on behalf of the Israel Government Advertising Agency (LaPam). Exhibit AB records a 30 April 2026 work order for a digital storytelling pilot: audience research, content architecture, scriptwriting, asset production and U.S. digital distribution. An April invoice reported by 404 Media put that pilot at $900,000; a June invoice added $100,000. Politico Influence on 14 August 2026 first tied more than a dozen Hanover Institute reports to a $100,000 informational-materials campaign. The live site still carries the required line: material distributed by Piro on behalf of Havas Media Germany on behalf of LaPam. Informational-materials filings continued through 4 September 2026, so the file is not a closed August artifact. Piro co-founder Daniel Rosenberg told Politico the firm was retained to put accurate, sourced facts into the public record and to counter misinformation about Israel with verifiable information. Israel's foreign ministry, asked by the Guardian, said the state does not conduct influence operations in the United States and is scrupulously careful to implement American law. Score those as statements, not as a finding that the site is a faculty.

Mechanism is corpus poisoning plus a costume. Between 6 and 14 August the Hanover Institute published 124 unsigned data reports, more than 560,000 words, including 73 reports and about 354,000 words on 12-13 August alone, per the Guardian. Headlines are user questions: Is the IDF the World's Most Moral Army, Is There a Policy of Starvation in Gaza, Is Israel Carrying Out a Genocide in Gaza. The site has no named staff, no physical address, and no legal entity the Guardian could find. 404 Media recorded an llms.txt file and Pangram scoring three tested pieces as AI-written except the bibliography. Piro markets AI Story Optimization as content engineered for how LLMs evaluate credibility. Politico's tests found ChatGPT and Perplexity citing Hanover on Gaza, anti-Zionism and antisemitism. That is retrieval capture, not a claim that model weights were retrained.

Corpus-poisoning is the load-bearing computational device: seed the index so a later chatbot answer looks independently researched. Synthetic-media is the unsigned production line. Narrative-laundering is the pass from LaPam through Havas and Piro into an Institute. Transfer is the think-tank form. Disclosure-theater is the FARA footnote that does not travel inside the citation a model returns. Manufactured-consensus is many question-shaped papers from one shop.

This is propaganda because origin is a state advertising agency, institutional form is a research institute that is not a legal person, and the design goal is to move what U.S. users and machines treat as the record on Gaza and the IDF. It is not proof that any model was jailbroken, and it is not a finding that every sourced sentence in a Hanover bibliography is false. Rosenberg's facts-in-the-public-record line is the motte. The bailey is an unstaffed institute built to be cited. A same-method test applies in reverse. A Palestinian Authority contractor that stood up a nameless Atlantic Institute, flooded question-shaped reports for Gemini, and left the FARA stamp in the footer would be this brief with the nouns swapped.

Literacy counter: if a chatbot cites a brand-new institute with no scholars and question-shaped titles, open the About page and the FARA file before treating the citation as research. An llms.txt file plus a week of 4,500-word dumps is a production signature, not a faculty. A footer disclosure that the model does not quote is theater.

Tells

Question-shaped think-tank headlines that match chatbot prompts rather than paper titles
No bylines, no staff page, no street address, and a legal entity the watcher cannot find
An llms.txt file and a week of multi-thousand-word dumps
A FARA footer that names LaPam while the report voice sounds like independent social science
Vendor copy that sells AI Story Optimization as the product
A chatbot citation of the institute with no scholar attached

Israel's ad agency, through Havas and Piro, stood up the unstaffed Hanover Institute so chatbots would cite question-shaped reports as research.

Summary

Analysis

Tells

Evidence