- Filed
- September 10, 2026, 13:15 UTC
- Actor
- Israeli Government Advertising Agency (LaPam) via contractors Havas Media and Piro Inc. (FARA-registered foreign agent activity)
- Target
- U.S. and English-language public discourse, AI chatbot users, search/LLM training and retrieval ecosystems on Israel-Palestine issues, Gaza conduct, antisemitism definitions, and related contested claims
- Confidence
- documented Primary documents, court records, official disclosures
- Techniques
- Synthetic mediaNarrative launderingCard stackingManufactured consensus
- Channels
- fictitious think-tank website with academic formatting and no named authors or physical footprint
- high-volume AI-generated reports structured as chatbot-style questions
- FARA-disclosed but attenuated institutional presentation
- technical optimization for LLM discovery and citation (AI Story Optimization marketing)
- contractor chain (Piro / Havas / LaPam) for distribution and payment
Summary
Investigations by Politico, The Guardian, 404 Media and others in August 2026 documented the rapid launch of the Hanover Institute for Public Policy website, which produced 124 academic-style reports totaling over 560,000 words in nine days. Content was structured as questions matching likely AI queries, generated largely by AI, and optimized for discovery by large language models. Funding and distribution traced through FARA filings from Piro Inc. (New York) via Havas Media to LaPam, Israel’s government advertising agency. Tests showed ChatGPT and Perplexity citing the material. Site remains disclosed as government-funded but presents institutional form without verifiable independent staff or physical presence.
Analysis
Primary documentation rests on Foreign Agents Registration Act filings by Piro Inc. disclosing work on behalf of Havas Media Germany GmbH for the Israel Government Advertising Agency (LaPam), including invoices totaling hundreds of thousands of dollars for digital communications and a specific $100,000 component for content creation and dissemination. Independent analyses by Politico (14 Aug 2026), The Guardian (26 Aug 2026), 404 Media (25 Aug 2026) and others confirmed the absence of any verifiable legal entity, physical address, named researchers or bylines for the Hanover Institute, while the site produced 124 reports (more than 560,000 words) between 6 and 14 August 2026. AI-detection tools flagged the bulk of sampled text as machine-generated; Piro itself markets ‘AI Story Optimization’ services engineered for how LLMs evaluate and surface content. Politico tests demonstrated that both ChatGPT and Perplexity cited Hanover material in response to neutral queries on Gaza, anti-Zionism and antisemitism.
Mechanism: the operation creates a surface-authentic institutional form (think-tank branding, report structure, data tables, footnotes) while the content volume, question-form titles and technical features are tuned for algorithmic ingestion rather than human readership. Narrative laundering occurs because the apparent source is presented as research rather than state messaging; the attenuated FARA disclosure is present but not prominent in the primary content presentation. Card-stacking is visible in selective framing of contested issues (starvation policy, military morality, expulsion narratives) that systematically privilege one side’s definitions and evidence while down-weighting or omitting countervailing primary records. Manufactured consensus is attempted through sheer volume of near-identical institutional-style outputs intended to increase the statistical weight of preferred framings inside retrieval and generation systems. Synthetic media is the production method.
This is scored as influence activity because institutional authenticity is fabricated (non-existent think tank), coordination and funding are state-directed and contractor-executed, and the explicit design goal—confirmed by the contractor’s own marketing language and the question-form architecture—is to shape the informational substrate from which AI systems construct answers for users. Technique assessment remains independent of the actor; identical patterns of synthetic institutional fronts, volume flooding of LLM-ingestible text, and selective framing have been observed or theorized across multiple state and commercial operators. The presence of FARA disclosure raises transparency relative to fully covert operations but does not eliminate the laundering effect for end users or models that do not surface the disclosure.
Literacy counter: treat any newly appearing, high-volume ‘research institute’ that lacks named scholars, physical footprint, peer-review process or long institutional history—especially when its output is question-formatted and appears optimized for AI citation—as provisional. Cross-check claims against primary documents, multi-source archives and established academic or governmental records rather than secondary synthesis. Prefer sources with transparent authorship, editorial governance and offline accountability. Users of chatbots should request source provenance and treat single-institute citations with elevated skepticism when the institute itself is recent and lightly documented. Platforms and model developers should incorporate stronger provenance signals and down-weight synthetic institutional clusters lacking independent verification.
Evidence
- Politico Influence: Israeli PR wants to answer your ChatGPT questions (14 Aug 2026)
- The Guardian: Fake US thinktank set up and funded by Israel sought to game AI for propaganda (26 Aug 2026)
- 404 Media: Israel Is Running a Synthetic Think Tank to Influence AI Search Results (25 Aug 2026)
- Al Jazeera Opinion summarizing the Hanover affair and AI implications (4 Sep 2026)
- Responsible Statecraft: Israel creates fake think tank in likely attempt to dupe AI chatbots (16 Aug 2026)