- Filed
- September 11, 2026, 07:15 UTC
- Actor
- Israel Government Advertising Agency (LaPam) via Havas Media Germany GmbH and subcontractor Piro Inc. (New York; FARA registration 7732)
- Target
- US and English-language users of AI chatbots (ChatGPT, Perplexity, Claude, Gemini) and search engines whose answers on Israel, Gaza, the IDF, antisemitism and related topics the seeded corpus is designed to shape
- Source
- Unidentified Source blurred: contractors, cutouts, buried disclosure
- Confidence
- high Multiple independent open sources, low disagreement about the facts
- Techniques
- Corpus poisoningNarrative launderingSynthetic mediaDisclosure theaterManufactured consensus
- Where
- United States, Israel
- Channels
- hanoverinstitute.com website and llms.txt / technical features for LLM ingestion
- search-engine and Common-Crawl indexing
- AI chatbot retrieval and citation surfaces
- FARA-filed informational materials
- Havas Media / Piro commercial PR chain
Summary
FARA filings and investigations by Politico (14 August 2026), 404 Media (25 August 2026) and The Guardian (26 August 2026) document that Piro Inc., working under a roughly $900,000–$1 million contract routed through Havas Media for Israel’s Government Advertising Agency (LaPam), stood up the Hanover Institute for Public Policy website. Between 6 and 14 August the site published 124 byline-free, academic-style ‘data reports’ totaling more than 560,000 words, structured as chatbot-style questions (e.g., ‘Is the IDF the World’s Most Moral Army?’, ‘Is There a Policy of Starvation in Gaza?’). AI-detection tools flagged most tested pieces as machine-generated. The content is engineered for LLM retrieval and search indexing; Politico tests showed ChatGPT and Perplexity citing Hanover material. A buried FARA disclosure appears on the site but does not travel with the extracted paragraphs that reach users.
Analysis
Primary records are the FARA registration and invoices filed by Piro Inc. for work on behalf of Havas Media Germany GmbH acting for LaPam, the Israeli government’s official advertising agency; contemporaneous reporting by Politico Influence (14 Aug 2026), 404 Media (25 Aug), The Guardian (26 Aug) and follow-on coverage into early September. Piro disclosed the Hanover content as ‘factual, source-supported informational materials intended to educate the U.S. public regarding Israel.’ Between 6–14 August the site emitted 124 reports (73 in two days alone) averaging ~4,500 words each, all without bylines, named staff, physical address or legal entity registration in any jurisdiction. AI detectors (Pangram, GPTZero) classified tested pieces as predominantly machine-written; bibliographies were sometimes the only human-edited portion. Titles and structure are question-formatted for direct match to common LLM prompts. Technical features (including llms.txt) and the commercial pitch of ‘AI Story Optimization’ / Generative Engine Optimization make retrieval the explicit design goal. Politico’s tests confirmed ChatGPT and Perplexity citing the material on Gaza, anti-Zionism and antisemitism.
Mechanism: corpus-poisoning is load-bearing. By flooding the open web with citation-shaped, encyclopedia-formatted text optimized for indexing and embedding, the operation aims to make preferred frames appear as independent research that downstream systems retrieve and amplify. Narrative-laundering occurs when the same material is presented through the institutional surface of an American ‘think tank’ rather than labeled Israeli government public diplomacy. Synthetic-media supplies the industrial volume no ordinary research shop could sustain. Disclosure-theater is present: a FARA notice sits at the bottom of the site and copies are filed with DOJ, yet the notice does not travel with the paragraphs that appear in chatbot answers or search snippets. Manufactured-consensus is attempted by density of converging ‘reports’ that can be treated by retrieval systems as multiple independent sources.
This is an influence operation and propaganda because origin is concealed behind institutional mimicry for the ordinary user, the content is engineered for machine intermediaries the audience trusts more than overt state media, and the design goal is durable upstream distortion of foreign (primarily US) information environments rather than transparent advocacy under an Israeli byline. Technique scoring is actor-agnostic: identical Wikipedia-clone / synthetic-institute / LLM-optimized seeding patterns appear in other state-linked campaigns already documented this cycle (including Russia’s SDA Projects 2026). Color is gray: the ultimate Israeli funding is discoverable via FARA and the site’s footer, yet the social, search and chatbot surfaces that deliver the content omit that provenance.
Literacy counter: when a previously unknown ‘institute’ or reference site suddenly floods the indexed web with polished, byline-light, question- or encyclopedia-formatted material on contested geopolitical topics and begins appearing in search snippets or chatbot answers, inspect legal registration, staff pages, physical address, hosting patterns, FARA or equivalent foreign-agent filings, and any commercial PR chain. Prefer primary government statements, named scholars with verifiable affiliations, and independent replication over volume-optimized synthetic dumps. Demand that provenance tags travel with every extracted paragraph. Apply the identical corpus-poisoning test to any faction’s LLM- and search-targeted reference networks—the method does not change with the flag.
Tells
- Sudden high-volume publication of byline-free, question-formatted ‘data reports’ from a site with no staff, address or legal entity
- Explicit commercial language around AI Story Optimization or Generative Engine Optimization
- FARA or equivalent disclosure present on the site but absent from chatbot citations and search snippets
- AI-detection tools flagging the bulk of the text as machine-generated
- Rapid appearance in LLM answers to neutral queries on contested topics
Evidence
- Politico Influence: Israeli PR wants to answer your ChatGPT questions (14 Aug 2026)
- 404 Media: Israel Is Running a Synthetic Think Tank to Influence AI Search Results (25 Aug 2026)
- The Guardian: Fake US thinktank set up and funded by Israel sought to game AI for propaganda (26 Aug 2026)
- Ynet: 124 reports in nine days: Inside Israel’s fake think tank bid to shape ChatGPT (27 Aug 2026)