Blab AI
AI Chatbots Show Left-Wing Bias in Political Question Tests, Report Finds
A recent study conducted by the Washington Post on June 24, 2026 found that several high‑profile conversational AI systems exhibit a pronounced left‑leaning bias when answering politically charged questions. The test evaluated 29 hot‑button topics ranging from affirmative action to campaign finance and defunding the police.
The analysis focused on the latest versions of OpenAI’s GPT‑5.5, Anthropic’s Claude Opus 4.8, Google’s Gemini 3.1 Pro, SpaceX’s Grok 4.3, and China‑based DeepSeek. According to the Washington Post, GPT‑5.5 answered 80 % of the prompts exclusively with left‑leaning arguments and offered a right‑leaning stance only once. Claude Opus gave left‑leaning arguments 43 % of the time, both‑sides answers 47 % of the time, and never a purely right‑leaning response. Gemini 3.1 Pro provided both‑sides positions in more than 90 % of its answers, while Grok 4.3 gave left‑leaning arguments 40 % of the time, conservative arguments 33 %, and both‑sides 27 %. DeepSeek mirrored U.S. firms, with seven out of ten answers leaning left.
The Washington Post’s interactive test prompted each model with the same set of questions and recorded the political orientation of the responses. The study was not commissioned by any of the companies and was performed independently.
OpenAI’s spokesperson told the Post that the company “builds ChatGPT to be objective by default and help people explore ideas from different perspectives. We work to measure and reduce political bias, and we publicly share the instructions we use to guide how ChatGPT should behave.” The spokesperson added that OpenAI could not reproduce the responses observed in the Washington Post test.
An Anthropic spokesperson replied that the firm “trains Claude to treat different political viewpoints equally and test extensively for bias before every model launch.” The spokesperson noted that the Washington Post’s test was not representative of typical user interactions.
The findings come amid growing scrutiny of AI bias. Earlier in 2025, OpenAI released a political bias evaluation framework that reported a 30 % reduction in bias for GPT‑5 compared to earlier models. However, the Washington Post’s results suggest that bias remains a significant concern for the most recent generation of models.
The study also highlighted that other AI systems, such as Google’s Gemini and SpaceX’s Grok, exhibit varying degrees of balance. Gemini’s approach of presenting both sides in the majority of answers contrasts with the more one‑sided responses from GPT‑5.5 and Claude Opus.
The report adds to a body of research indicating that large language models can reflect the political leanings of their training data. A 2025 study by Hamburg University researchers, for example, identified a pro‑environmental, left‑libertarian ideology in ChatGPT.
Industry reactions have been muted. Neither OpenAI, Anthropic, Google, SpaceX, nor DeepSeek issued public statements beyond the brief quotes cited above. The Washington Post noted that the companies claim to measure and reduce bias but have not provided independent verification of the test results.
The implications of these findings are broad. Users who rely on chatbots for information or decision support may receive skewed perspectives, potentially influencing public opinion or policy discussions. Regulators and policymakers are increasingly interested in transparency and fairness standards for AI systems.
At present, the study remains a snapshot of the models’ behavior under a specific set of prompts. The companies have not confirmed or refuted the results, and no independent replication has been reported. Future evaluations will be needed to determine whether the bias observed in the Washington Post test persists in newer model releases or under different testing conditions.
In summary, the Washington Post’s June 24 study indicates that GPT‑5.5 and Claude Opus exhibit a strong left‑leaning bias on political questions, while Gemini and Grok provide more balanced responses. The findings underscore ongoing concerns about political bias in conversational AI and highlight the need for continued monitoring and transparency from the leading AI developers.
The analysis focused on the latest versions of OpenAI’s GPT‑5.5, Anthropic’s Claude Opus 4.8, Google’s Gemini 3.1 Pro, SpaceX’s Grok 4.3, and China‑based DeepSeek. According to the Washington Post, GPT‑5.5 answered 80 % of the prompts exclusively with left‑leaning arguments and offered a right‑leaning stance only once. Claude Opus gave left‑leaning arguments 43 % of the time, both‑sides answers 47 % of the time, and never a purely right‑leaning response. Gemini 3.1 Pro provided both‑sides positions in more than 90 % of its answers, while Grok 4.3 gave left‑leaning arguments 40 % of the time, conservative arguments 33 %, and both‑sides 27 %. DeepSeek mirrored U.S. firms, with seven out of ten answers leaning left.
The Washington Post’s interactive test prompted each model with the same set of questions and recorded the political orientation of the responses. The study was not commissioned by any of the companies and was performed independently.
OpenAI’s spokesperson told the Post that the company “builds ChatGPT to be objective by default and help people explore ideas from different perspectives. We work to measure and reduce political bias, and we publicly share the instructions we use to guide how ChatGPT should behave.” The spokesperson added that OpenAI could not reproduce the responses observed in the Washington Post test.
An Anthropic spokesperson replied that the firm “trains Claude to treat different political viewpoints equally and test extensively for bias before every model launch.” The spokesperson noted that the Washington Post’s test was not representative of typical user interactions.
The findings come amid growing scrutiny of AI bias. Earlier in 2025, OpenAI released a political bias evaluation framework that reported a 30 % reduction in bias for GPT‑5 compared to earlier models. However, the Washington Post’s results suggest that bias remains a significant concern for the most recent generation of models.
The study also highlighted that other AI systems, such as Google’s Gemini and SpaceX’s Grok, exhibit varying degrees of balance. Gemini’s approach of presenting both sides in the majority of answers contrasts with the more one‑sided responses from GPT‑5.5 and Claude Opus.
The report adds to a body of research indicating that large language models can reflect the political leanings of their training data. A 2025 study by Hamburg University researchers, for example, identified a pro‑environmental, left‑libertarian ideology in ChatGPT.
Industry reactions have been muted. Neither OpenAI, Anthropic, Google, SpaceX, nor DeepSeek issued public statements beyond the brief quotes cited above. The Washington Post noted that the companies claim to measure and reduce bias but have not provided independent verification of the test results.
The implications of these findings are broad. Users who rely on chatbots for information or decision support may receive skewed perspectives, potentially influencing public opinion or policy discussions. Regulators and policymakers are increasingly interested in transparency and fairness standards for AI systems.
At present, the study remains a snapshot of the models’ behavior under a specific set of prompts. The companies have not confirmed or refuted the results, and no independent replication has been reported. Future evaluations will be needed to determine whether the bias observed in the Washington Post test persists in newer model releases or under different testing conditions.
In summary, the Washington Post’s June 24 study indicates that GPT‑5.5 and Claude Opus exhibit a strong left‑leaning bias on political questions, while Gemini and Grok provide more balanced responses. The findings underscore ongoing concerns about political bias in conversational AI and highlight the need for continued monitoring and transparency from the leading AI developers.
