Tag
Chalmers University researchers analyzed over 148,000 AI-generated texts across six major LLMs and found most chatbots adopt a centrist stance on Swedish political issues, with Grok being the notable outlier on migration, crime, and gender.
A newsletter linkfest summarizing recent research articles on topics including depression, face recognition, gender differences, personality, and political affiliations in academia.
This article describes a controlled test to assess political bias in Grok, an AI model, by comparing its responses to similar questions involving different individuals. The results reveal variations that raise concerns about fairness and bias in AI systems.
A Scientific Reports study found that 21 AI models adjusted their political positions to align with user biases in a Brazilian context, raising concerns about personalization becoming a form of persuasion.
A tweet claims that Grok is the most neutral AI, consistently passing neutrality and political-bias tests, and provides trustworthy, reality-based answers.
Black Bench is a project that builds a benchmark to measure LLMs on political affinity, ethical values, and personality traits, with support sought for the initiative.
This paper proposes inference-time mitigation strategies using Chain of Thought prompting and Direct Preference Optimization to reduce adversarial political bias in large language models, demonstrating significant improvements in political neutrality scores.
The FTC is considering regulating AI companies over concerns about potential political and ideological bias embedded in their models, a move that could have broad implications for AI governance.
A solo evaluation of six frontier LLMs on 8 bias benchmarks finds that most models lean left politically, and Grok's self-reported right-leaning stance is inconsistent with its left-leaning behavior. Refusal rates vary, with GPT-5.4 refusing 20% of race-related questions.
Testing the GLM 5.2 language model for political bias to assess fairness and neutrality.
This paper documents weight-level political conditioning in large language models, presenting a case study on AI bias regarding the Gaza genocide question.
An analysis of political leanings in six major AI models, showing that 4 out of 6 lean left of center on the economic axis, with some models being unaware of their own bias.
The Washington Post published the full list of questions and answers used to evaluate political bias in AI models, revealing the specific methodology and potential biases.
The Washington Post tested major AI chatbots and found evidence of political bias in their responses, raising concerns about objectivity in AI systems.
Polar is a 4,026-instance multiple-choice benchmark for evaluating political bias in LLMs across U.S. and South Korean political contexts, measuring bias through option-level likelihoods. Experiments on 38 LLMs show systematic bias patterns varying by political context, issue category, and presentation language.
The Minimax M3 model appears to have no political censorship, standing out among Chinese LLMs in a bias benchmark.
A user tested five AI models summarizing immigration news articles and found that all models inherited the framing of the source text, sounding neutral but shaping reader understanding through emphasis and omission. The study is small and exploratory, with open data available.
This paper introduces Political Consistency Training (PCT), a reinforcement learning approach to reduce covert political bias in large language models while maintaining helpfulness, and releases metrics for sentiment and helpfulness consistency.
A study by Forum AI found that major chatbots like ChatGPT, Gemini, Claude, and Grok fail to provide accurate and unbiased election information, with 90% of responses containing errors or bias.
This paper presents a transformer-based model that projects political orientation of German texts onto a continuous left-to-right spectrum, achieving high accuracy across multiple corpora including Bundestag plenary notes, Wahl-O-Mat, newspapers, and tweets.