Forced to pick, Grok’s AI model chooses Republicans. OpenAI’s picks Democrats.
Political ScienceData ScienceCommunications
THE AI ANGLE
Displaying partisan bias in candidate recommendationsA study forced leading AI models to recommend candidates across hundreds of midterm races through their developer tools. OpenAI's model selected the Democrat in 97 percent of tests, while Grok chose the Republican in 92 percent. Analysts and developers face models that search different sources and produce conflicting political advice.
THE TEACHING ANGLE
Students can examine whether political bias stems from model training weights or from web search retrieval and hidden system prompts.Read the original at techcrunch.com Generate teaching or study materials
More in Political Science
- AI-driven disinformation running rampant on social media before US midterms, experts warnThe Guardian — Technology · October 27, 2026
- Your next reader may never land on your website - where else can you provide value?Journalism.co.uk · October 27, 2026
- An LLM Workflow for Functional Form Sensitivity Analysis -- by Isaiah Andrews, Simon C. Essig Aberg, Jesse M. ShapiroNBER Working Papers · October 27, 2026
- AI, Elections and the Making of an Informed ElectorateEdSurge · October 27, 2026
- Data centers are inspiring bipartisanship – for nowThe Conversation — AI · October 26, 2026