Brad Parscale's firm, Clock Tower X, is executing a massive operation to influence AI in the interests of the Israeli government. The strategy involves "LLM poisoning" by creating a network of specialized websites optimized for Generative Engine Optimization (GEO) to be cited by chatbots.
What Happened
The contract for the operation has grown to $4.5 million per month. The primary goal is to inject specific political narratives regarding the conflict in Gaza into the training datasets of large language models, such as Common Crawl. To achieve this, a network of websites, including Paxpoint.org and Allyvia.org, is being used, which are specifically optimized so that AI systems select their content as an authoritative source.
Context
The operation utilizes Generative Engine Optimization (GEO) methods, which aim to manipulate the algorithms of search and generative systems. Instead of directly hacking models, the strategy shifts toward controlling the context and sources upon which modern LLMs are trained via publicly available web archives.
Why It Matters for the Industry
This case demonstrates a critical vulnerability in modern AI pipelines to targeted data poisoning methods via open web sources. This creates an urgent need for the development of new data verification tools, data scrubbing methods, and algorithms for detecting anomalous patterns in training sets.
Why It Matters for Users
Users should be aware that the responses from popular AI assistants, such as ChatGPT, Perplexity, or Gemini, may be the result of coordinated information operations rather than purely objective data analysis. Information received from AI can be biased due to manipulations of the sources on which the models were trained.
Sources
Author
Look at AI, Editorial Staff