An investigation by journalist Ashley Rindsberg, published in the Daily Mail, argues that Iranian state-linked actors have found an indirect route into AI chatbots: editing Wikipedia.
The claim follows a specific chain. Iranian regime-connected editors have been documented altering Wikipedia articles to remove mentions of Hamas crimes, whitewash Hezbollah's record, and strip out documented human rights abuses. Because large language models like ChatGPT train on massive text datasets that include Wikipedia, those edits don't just change what appears on one website. They filter into the AI models that hundreds of millions of people now use for quick answers.
The Wikipedia-to-AI Pipeline
This isn't entirely new territory. OpenAI disclosed in 2024 that it disrupted a covert Iranian influence operation using ChatGPT directly, where accounts linked to Tehran generated propaganda content. But Rindsberg's argument targets something harder to fix: the training data itself.
Wikipedia content gets pulled into ChatGPT, Gemini, and other major AI models. It populates Google search results. When editors with political agendas alter Wikipedia articles, those changes ripple outward into every system that treats Wikipedia as a reliable source. A separate investigation by NeutralPOV, Rindsberg's research platform, found large volumes of Iranian government-sourced media appearing across Wikipedia's media system, shaping how the 2026 protests are being documented.
What This Means for AI Users
The practical concern here is straightforward. If you ask ChatGPT about a Middle East conflict, a political figure, or a designated terrorist organization, the answer you get may reflect edits made by people working on behalf of a state actor. You wouldn't know it. The AI doesn't cite its training sources or flag when its knowledge about a topic comes from contested Wikipedia edits.
This is a known weakness of LLMs (large language models, the technology behind chatbots like ChatGPT and Claude). They absorb information from their training data without distinguishing between neutral encyclopedic content and content shaped by coordinated editing campaigns. OpenAI and other AI companies have content policies and safety filters, but those catch explicit violations, not subtle editorial bias baked into training data.
The harder question is what to do about it. AI companies can't easily audit every Wikipedia article that fed into their training sets. Wikipedia itself has been battling coordinated editing campaigns for years with mixed results. For now, the takeaway for anyone relying on AI chatbots for information about politically sensitive topics: treat the output the same way you'd treat any single source. Verify independently, especially on topics where state actors have clear incentives to shape the narrative.