Study Finds Significant Political Bias in Grok and GPT AI Models
A new study by Convovo, led by Cornell University professor Tapan Parikh, reveals stark political leanings when AI models are forced to recommend political candidates via API. When prompted to choose a single candidate without hedging, SpaceXAI’s Grok recommended Republicans in 92% of cases, while OpenAI’s GPT model selected Democrats in 97% of instances across nearly 500 races. The research highlights a discrepancy between consumer-facing applications, which typically refuse to provide voting recommendations, and the underlying APIs used by developers. While OpenAI maintains its models are designed for objectivity, researchers found that both models exhibited implicit biases through their information sourcing and topic prioritization—GPT focusing on healthcare and abortion, while Grok prioritized immigration and taxes.
Topics: AI bias election integrity LLM research political neutrality