OpenAI’s newly released GPT-5.2 model, touted as its “most advanced frontier model for professional work,” is facing scrutiny after tests revealed it relied on the controversial online encyclopedia Grokipedia for information on sensitive topics. The findings, reported by the Guardian, raise questions about the sources powering the cutting-edge AI.
Questionable Sources in advanced AI
The report details instances where ChatGPT, powered by GPT-5.2, cited Grokipedia when discussing allegations about the Iranian government’s connection to MTN-Irancell, and when addressing questions surrounding Richard Evans, a British historian who served as an expert witness in a libel case involving Holocaust denier David Irving. Interestingly, the Guardian found ChatGPT did not utilize Grokipedia when responding to prompts about media bias concerning Donald Trump.
- Grokipedia, created by xAI, has previously been criticized for including citations to neo-Nazi forums.
- A US-based study indicated the AI-generated encyclopedia cited sources deemed “questionable” and “problematic.”
Specifically, ChatGPT drew upon Grokipedia when addressing allegations about the iranian government’s connection to MTN-Irancell, and when discussing questions surrounding Richard Evans, a British historian who served as an expert witness in a libel case involving Holocaust denier David Irving. This reliance on a relatively new and unvetted source is raising eyebrows among experts.
What is Grokipedia? The online encyclopedia, developed by xAI, launched prior to the release of GPT-5.2 in December. Though, it quickly attracted controversy after being found to include citations to extremist websites, prompting investigations by authorities.
A study by US researchers revealed that Grokipedia frequently cited sources considered “questionable” and “problematic,” raising concerns about the quality and reliability of its information.
OpenAI responded to the Guardian‘s report by stating that GPT-5.2 searches the web for a “broad range of publicly available sources and viewpoints,” while also implementing “safety filters to reduce the risk of surfacing links associated with high-severity harms.” The company did not directly address the specific instances of Grokipedia being cited.
The GPT-5.2 model was released in December with the aim of improving performance on professional tasks, such as spreadsheet creation and complex data analysis. The revelation that it may be drawing information from a source with a questionable track record raises important questions about the safeguards in place to ensure the accuracy and impartiality of AI-generated content.
