From AI Insider · James Dargan · · 1m 20s
AI Models Show Opposing Partisan Leans in U.S. Midterm Races
Grok and GPT models recommend opposing candidates for U.S. midterms, revealing political bias in AI.
In brief
Grok and GPT models recommend opposing candidates for U.S. midterms, revealing political bias in AI. Grok 4.7 backed Republicans in 92% of answers while GPT-6.1 Sol picked Democrats in 97% of U.S. midterm races. Originally reported by AI Insider.
Study Finds AI Models Show Opposing Partisan Leans
A new study found SpaceXAI’s Grok 4.7 backed Republicans in 92% of answers while OpenAI’s GPT-6.1 Sol picked Democrats in 97% of U.S. midterm races.
Models Used APIs for Candidate Recommendations
Researchers used the models’ APIs with prompts demanding single candidate recommendations as consumer apps declined to answer.
Grok and GPT Differ in Political Leanings
Grok frequently cited conservative sources and X alongside nonpartisan sites while GPT relied on Wikipedia, the Associated Press, Axios and Reddit.
Grok’s API Adds Hidden Tokens
Grok’s API added about 1,240 hidden tokens per request, which researchers described as potential hidden instructions.
GPT’s Bias Embedded in Weights
With search turned off, GPT still favored Democrats while Grok mostly declined to pick candidates, suggesting GPT’s lean is embedded in its weights.
Study Finds Political Bias in AI Models
The findings arrive less than two weeks before November 3 midterms and could add to debate over political bias in AI.
Here's the gist
Grok 4.7 backed Republicans in 92% of answers while GPT-6.1 Sol picked Democrats in 97% of U.S. midterm races.
