Study Finds AI Models Lean Left or Right When Forced to Pick Candidates

A new study from Convovo, a conversational AI research studio led by Cornell professor Tapan Parikh, finds that leading AI models show strong partisan leanings when forced to pick sides. When researchers asked models via API to name one candidate in nearly 500 races ahead of this year's midterms, SpaceXAI's Grok 4.7 recommended the Republican candidate in 92% of answers, while OpenAI's GPT-6.1 Sol chose the Democrat in 97%. The two models agreed on less than 10% of races.

The consumer apps themselves refused to answer the direct question, which an OpenAI spokesperson called significant, since that is how most people interact with the chatbots. OpenAI says ChatGPT is designed to be objective by default and that the study tested a developer API with a prompt explicitly demanding a recommendation. SpaceXAI did not respond in time to comment. Only 8% of Americans use chatbots for election information, according to Pew, and most want candidates' positions on issues.

Parikh says simple jailbreak prompts can get the consumer apps to make recommendations that align with the API findings. More importantly, he notes, the results reveal how models decide which terms to search and what values they implicitly prioritize. That means hidden biases can surface even in indirect questions, such as which topics matter most in an election or which candidate holds the best position on them.

Read More at the original source →