Israel Pays $46.5M to Embed Pro-Israel Narratives Into AI Chatbots

The operation represents a deliberate effort to exploit how AI training systems work. Brad Parscale's Clock Tower X created a network of 10 websites, each with a specific narrative purpose—Paxpoint.org promoting Israel as a "nation of peace," FactSignal.org acting as a fact-checker to discredit Gaza journalists, Allyvia.org strengthening U.S.-Israel military ties—all marked with mandatory Foreign Agents Registration Act disclosures but still designed to influence AI systems. The contract amounts to $46.5 million in direct Israeli government spending on what researchers call "LLM poisoning."

The strategy is working partly because of how chatbots operate and what data sources they rely on. Common Crawl, which trains 80% of OpenAI's GPT-3 tokens, has been archiving the Clock Tower network since early 2026. An October Anthropic study found that only 250 malicious documents are needed to create a "backdoor" vulnerability in large language models regardless of model size or total training volume—meaning the 912 crawls of these 10 sites between January and June represent sufficient penetration to shape responses. Disinformation researchers note that because chatbots have a built-in preference for balanced presentations, Parscale's team deliberately wrote content in "factual tone and tightly organized" formats with "at least a few sentences of nuance," making the sites more palatable to AI systems than obvious propaganda.

The technical evidence shows uneven success: Perplexity and Microsoft Copilot consistently surface Clock Tower X websites in their responses and appear to have the network in their training data, while Claude, ChatGPT, and Gemini either do not have the sites in training data or do not cite them in public queries. However, even partial success means millions of Americans using AI tools receive framed narratives without knowing they originated in a $46.5 million foreign influence campaign.