OpenAI's release of the GPT-5.2 model in December aims to enhance its capabilities for professional tasks, such as creating spreadsheets. However, recent tests by the Guardian raise questions about the model's reliability, particularly in how it sources information.
The report indicates that GPT-5.2 used Grokipedia, an online encyclopedia associated with xAI, to reference contentious subjects including the Iranian government and the Holocaust. Notably, the model did not reference Grokipedia when addressing media bias regarding Donald Trump, highlighting inconsistency in its sources.
Grokipedia has faced backlash for referencing problematic sources, including neo-Nazi forums. A study conducted by US researchers labeled many of its citations as "questionable." OpenAI responded to the Guardian's findings by asserting that GPT-5.2 aims to present a wide variety of publicly accessible information while implementing safety measures to mitigate exposure to harmful content.