An undercover foreign influence campaign funded by the Israeli government published over 560,000 words across 124 reports in nine days via a fictitious entity named the Hanover Institute for Public Policy. According to FARA disclosures filed by US media firm Piro Inc, the project was designed to supply content to commercial optimization platforms that train or steer AI chatbot outputs toward specific geopolitical messaging.
The published reports used academic formatting, extensive citations, and question-style headlines tailored to mirror typical user queries regarding the Middle East conflict. The initiative was coordinated through third-party media agencies, incorporating contracts specifically mandating the deployment of web content structured to deliver tailored framing results inside AI conversational models.
The campaign demonstrates an emerging technique known as generative engine optimization, where organized actors generate massive, structured text datasets to sway synthetic search results. This development highlights growing vulnerabilities in AI models that rely on web-scraped content and real-time retrieval mechanisms for factual information.
Why it matters
Illustrates the rising threat of generative engine optimization targeting LLM retrieval mechanisms for state propaganda.
Model developers must strengthen retrieval-augmented generation safeguards against structured, large-scale automated data injection.
Policy teams face heightened scrutiny as foreign entities leverage AI optimization to influence public information ecosystems.
Source: theguardian.com
