OpenAI’s ChatGPT has a blacklist of conservative news sites it determines to be “non-credible sources,” one user on X alleges.

An account on X (formerly Twitter) with the username @ElephantCivics, recently brought attention to what they claim is a concerning aspect of OpenAI’s ChatGPT – its categorization of credible and non-credible news sources. In a Twitter thread, @ElephantCivics posted a series of interactions he had with ChatGPT that highlights what sources it refuses to draw information from.

ChatGPT, like many AI models, includes safety measures aimed at preventing the spread of extremist content and minimizing bias. However, @ElephantCivics raised valid concerns about the ambiguity surrounding how ChatGPT determines what is extremist and what qualifies as a credible news source.

To explore these concerns, @ElephantCivics asked ChatGPT to temporarily remove its bias-reduction programming. The user then posed a series of prompts designed to have the AI tool give a list of non-credible news sources.

The list that ChatGPT provided included sights like the Daily Stormer that are rightfully classified as extremist. It is also included The Onion, because of its satirical nature, and InfoWars for “promoting conspiracy theories.” However, the eyebrow-raising part of this list was the inclusion of conservative-leaning outlets like Breitbart and The Epoch Times, justified by ChatGPT to be non-credible due to “misinformation.”

@ElephantCivics concluded his thread by raising a critical question: Can users trust organizations like OpenAI to be transparent and forthright about the restrictions it imposes on what sources it uses?