A newly created think tank, the Hanover Institute for Public Policy, has published over 100 AI-generated articles in less than a month as part of a campaign funded by Israel to influence the results of AI-powered search engines and chatbots, according to a report from 404 Media. The effort, first flagged by Politico, represents a novel and concerning use of automated propaganda to exploit the way large language models (LLMs) train on web content.
The Hanover Institute for Public Policy, launched in late July 2026, operates as a front organization bankrolled by the Israeli government and run by the American advertising firm Piro Inc. According to a report from 404 Media, the Institute's website has published over 100 articles in less than a month, with each piece framed around a leading question designed to sway public perception on topics such as Palestinian land claims, accusations of Israeli aggression, and allegations of starvation policies in Gaza.
Analysis by the AI detection tool Pangram found that three articles tested from the Institute were almost entirely generated by artificial intelligence, with only the bibliographies appearing to be human-written. The site's images are also AI-generated. The articles lack bylines, a practice the Institute defends on its 'About' page, stating that findings should stand on sources rather than authors, and that anonymity protects researchers from harassment. The pages do, however, disclose that the site is registered with the U.S. Department of Justice.
The strategy appears tailored to exploit a weakness in current large language models, which constantly scrape the open web for data to inform their responses. By flooding the internet with authoritative-sounding, AI-optimized content, the Institute aims to seed chatbot training data with pro-Israel narratives, effectively turning the world's most advanced AI systems into unwitting propagators of its messaging.
The content itself has been described as dense and often nonsensical. In one article discussing claims of restricted food access in Gaza, the text reads: "A reader who has encountered the claim as a consensus has encountered something real. The reframe is the denominator. Six documents from five bodies is a count of assertions, not a change in the kind of evidence available." This style, while confusing to human readers, is crafted to pass the statistical pattern-matching tests of LLMs, reinforcing certain claims by sheer volume of algorithmic approval.
Experts in information warfare have expressed alarm, noting that this tactic could be a harbinger of a new era in propaganda. Unlike traditional disinformation campaigns that target human audiences, this effort explicitly targets the AI systems that billions of people now rely on for information. The operation is particularly hard to counter because the content is not overtly false in a verifiable way; instead, it creates a skewed statistical weight in the training data that tilts chatbot outputs toward a specific viewpoint.
The Israeli government has not commented directly on the operation, but the disclosure of U.S. Department of Justice registration suggests it may be considered a legal form of foreign lobbying. Critics argue that it represents a deceptive and manipulative use of technology to circumvent independent journalism and fact-checking.
Analysis
Why This Matters
- This marks a dangerous evolution in information warfare: propaganda is no longer just for human consumption but is being precisely engineered to manipulate the AI models that millions use daily for news, research, and decision-making.
- If successful, similar campaigns could become the norm, allowing well-funded governments and corporations to effectively buy influence over AI-generated knowledge, undermining the reliability of these tools as objective sources of information.
- The tactic is difficult to detect and counter, as it leverages the core mechanism of LLM training (mass web scraping) rather than relying on easily disputed falsehoods, creating a 'Ghost in the Machine' for the AI age.
Background
For decades, governments and political groups have engaged in astroturfing—creating fake grassroots campaigns or think tanks to influence public opinion. However, the prior model depended on manipulating human journalists and social media algorithms, which required convincing or deceiving people.
The rapid adoption of large language models like ChatGPT and Google Gemini has created a new vector. These models are trained on vast swaths of the internet, with a premium placed on content that appears factual, well-sourced, and authoritative. The Hanover Institute's strategy, first exposed by Politico and deepened by 404 Media, is a direct response to this new reality. By generating thousands of pages of AI-written, question-shaped content, the Institute is effectively trying to poison the well of AI training data with its preferred narratives. The use of an AI detection tool to confirm the content's origin highlights the meta nature of this conflict: AI is being used to both create and detect the propaganda.
Key Perspectives
[Pro-Israel advocates and the Institute's defenders]: The Hanover Institute's approach may be framed as legitimate public diplomacy and advocacy in the digital age. The disclosure of DOJ registration suggests a belief in legal compliance. Supporters would argue that in an era where anti-Israel sentiment is widespread online, and where AI systems often rely on biased open-source data, this is a necessary corrective to ensure that Israel's case is heard.
[Critics and disinformation researchers]: They view this as a clear and present danger to information integrity. Critics argue that the lack of transparency (no named authors, AI-generated content) is deliberately deceptive. The content is designed not to inform, but to game algorithmic systems, representing a violation of trust in both journalism and technology. The sheer volume and nonsensical nature of the text demonstrate that its purpose is not human communication but AI manipulation.
[Policymakers and tech companies]: This situation puts pressure on AI developers and platforms to find ways to detect and mitigate such 'synthetic' influence campaigns. It raises questions about the legal framework for foreign-funded AI propaganda, and whether current lobbying laws adequately cover such indirect manipulation of algorithmic outputs.
What to Watch
- Detection and response from major AI companies like OpenAI and Google. Will they update their training data filters to exclude content from known synthetic think tanks, or will the arms race continue?
- The scale of the problem. If this tactic proves effective, expect a flood of similar operations from various states and interest groups, potentially overwhelming the ability of AI systems to learn from reliable sources.
- Legal and regulatory action. The U.S. Department of Justice and other bodies may scrutinize whether this constitutes undisclosed foreign propaganda, leading to potential fines or restrictions on the operators.