Study finds Grok 4.7 and GPT-6.1 Sol show partisan biases
A new study says not all AI is neutral. Some models actually have strong political leanings.
SpaceXAI's Grok 4.7 backed Republican candidates in 92% of its picks, while OpenAI's GPT-6.1 Sol went for Democrats 97% of the time.
Out of nearly 500 election races, the two agreed on less than 10%, showing just how different their recommendations can be.
API tests reveal partisan source preferences
The study found these political preferences in API testing, while the consumer apps initially refused to answer.
Grok often links to conservative sites like Turning Point Action, while GPT prefers sources like Wikipedia and AP.
Even their hot topics differ: GPT talks healthcare and abortion, Grok focuses on immigration and crime.
As elections get closer, researcher Tapan Parikh points out that AI models are not just neutral tools; they carry value systems that can influence decision-making, raising big questions about how much sway AI might have over what we think and vote for.