NewsGuard, a media trustworthiness evaluator, introduced an AI chatbot this week that asserts gender identities may range from 72 to infinite and rejects the concept of binary biological sex, per a Washington Free Beacon review. The tool, marketed as a "personal fact-checker," has been endorsed by The Atlantic and Snopes but draws criticism for omitting moderate or conservative viewpoints.
AI Fact-Checker's Stance on Gender and Sex
When asked about biological sex, the chatbot dismissed the "traditional two-sex framework" as overly simplistic and cited a Huffington Post essay by a gender studies professor but excluded a Wall Street Journal op-ed by biologists arguing sex is "functionally binary." It referenced a MedicineNet article listing 72 gender identities, separating gender from anatomy. In contrast, Anthropic's Claude stated, "The standard biological answer is two, defined by gamete type," while acknowledging complexity in physical expression.
Bias Concerns in Source Selection
The chatbot's design, pulling only from sources rated above 60 on NewsGuard's reliability scale, creates a leftward skew. A study found right-leaning outlets average a 26.4 trust score (below the threshold) versus 60.6 for left-leaning sites. This could exclude conservative perspectives from AI training data, institutionalizing ideological biases. Matt Skibinski, NewsGuard's COO, admitted the tool "isn't perfect" but defended its accuracy on factual matters.
Testing revealed errors, such as misidentifying Dan Morenoff, executive director of the American Civil Rights Project, as "Jeffrey Morenoff," a sociologist. The chatbot also botched dates of articles and conflated UCLA administrators. NewsGuard, co-founded by Steven Brill and Gordon Crovitz, sells data to advertisers and AI firms to combat "misinformation," though no major AI company has yet licensed its domestic source ratings.
Skibinski clarified that the chatbot uses a "commercially available LLM" restricted to non-paywalled articles from vetted sources. Critics argue this approach risks embedding NewsGuard's biases, such as rating Chinese state media higher than some conservative outlets, into AI systems framed as neutral. The company now distances itself from earlier claims about training AI models with its data, stating, "I don't really see how our data would be used for model training."