Source: 404 Media
A state-funded organization has begun seeding AI training datasets with over 100 articles in weeks, targeting language models' appetite for fresh text to shape their output on Israeli policy. This marks the first documented case of a government building infrastructure to manipulate AI outputs at scale—not through API prompts or user-facing tactics, but by poisoning source material during the training and fine-tuning phases. The strategy exploits how AI companies remain largely indiscriminate about training data origins, treating institutional-sounding publications as credible sources regardless of actual editorial independence.