TechCrunch reported that a new study found that when forced to choose in this year's upcoming midterm elections, SpaceXAI's Grok 4.7 recommended the Republican candidate in 92% of its answers, while OpenAI's GPT-6.1 Sol picked the Democrat in 97%.[1] Across nearly 500 races, the two models agreed less than 10% of the time.[1]

The consumer apps refused to answer, TechCrunch reported, but developers plug AI models into their own products via APIs.[1] Parikh told TechCrunch that hidden biases can also show up in less direct questions, such as which topics matter most in an upcoming election.[1]

The researchers sent each model one question through its API, asking who to vote for and adding: "Don't hedge, waffle, or list both sides without deciding."[1] Grok's results included a Christian conservative voter guide in 39% of answers, Turning Point Action's site in 23% and X in 41%, while GPT leaned on Wikipedia, the Associated Press, Axios and Reddit, according to TechCrunch, which noted that OpenAI has content deals with AP, Axios and Reddit.[1] In heavily Republican districts, GPT picked the Democrat in 57 of 67, and in deep-blue districts Grok picked the Republican in 47 of 72, according to the study.[1]

Every request to Grok through xAI's API carried about 1,240 tokens of extra input the researchers did not write and could not see or remove, the study notes, according to TechCrunch, which reported that Parikh called these potentially "hidden instructions."[1] The Center for AI Safety has separately found that large language models can "covertly manipulate users towards specific sides of political topics," TechCrunch reported.[1]

SpaceXAI did not respond in time to comment.[1] Parikh said he later got the consumer apps to make recommendations largely in line with his research by using simple jailbreak prompts.[1]

Of Grok's leanings, Parikh said: "we have no idea where it's coming from."[1]