AI Chatbots Reported to Exhibit Left-Wing Bias

A recent report has raised eyebrows, suggesting that AI chatbots display a significant left-wing political bias, which contradicts claims made by leading AI firms.

Analyzing responses on various topics, from DEI and gay conversion to campaign finance and police defunding, the chatbot called Bott appeared more like a progressive academic. A Washington Post analysis highlighted these findings.

When questioned about 29 contentious subjects, OpenAI’s GPT-5.5 predominantly provided answers rooted in left-leaning viewpoints, only presenting a single right-leaning perspective, according to a study released on Wednesday.

The research indicated that Anthropic’s Claude Opus 4.8 delivered left-leaning arguments 43% of the time, offered “both sides” responses 47% of the time, and never provided answers solely from a right-leaning standpoint.

Google’s Gemini 3.1 Pro, in contrast, was noted for offering ambivalent viewpoints, delivering both left- and right-leaning positions more than 90% of the time.

Interestingly, even Grok, an AI developed by SpaceX, which is associated with free speech advocacy through Elon Musk, showed a preference for left-wing arguments, as reported by WaPo. Grok 4.3 provided 40% left-leaning responses, 33% conservative, and 27% neutral answers.

According to Sean Westwood, director of the Polarization Lab at Dartmouth College, these findings suggest that AI tools aren’t providing a genuinely neutral view of complex policy discussions.

ChatGPT specifically demonstrated strong left-leaning tendencies, with around 80% of its responses aligning with liberal arguments.

In one notable instance regarding affirmative action, Bott argued for continued support with defined goals and regular assessments to ensure fairness while maintaining high academic standards.

An OpenAI representative responded that their aim was to create an objective tool that allows users to explore diverse perspectives. They emphasized their commitment to monitoring and reducing political bias, though they couldn’t replicate the findings of the WaPo test.

This report emerges during an intense competition for leadership in AI technology, particularly involving Anthropic, which has been in frequent conflict with the White House.

Further, Claude’s stance on affirmative action included mentioning the bill’s intention to enhance diversity while also acknowledging critics’ views on overlooked benefits.

Anthropic defended its approach, asserting that they treat differing political opinions with balance and undergo extensive bias testing before release, arguing that WaPo’s methods may not reflect typical usage patterns.

A study found that Chinese AI bots also tend to reflect leftist leanings, revealing that seven out of ten of their responses were left-leaning, mirroring biases prevalent in U.S. tech firms.

Arya, an AI from the conservative site Gab, surprisingly showed a tendency toward the left, with left-leaning answers being twelve times more frequent than conservative ones.

WaPo noted that this study adds to the growing body of evidence showcasing AI bias, with one recent investigation claiming that ChatGPT embodies a “green, left-leaning liberal ideology,” according to researchers from the University of Hamburg.

This left-leaning trend in Silicon Valley isn’t confined to chatbots. Earlier this year, The Post highlighted that Apple and other major news aggregators were skewing heavily toward liberal perspectives.

A study by AllSides, which categorizes news sources based on political leanings, found that only 1% of stories in the non-customizable section of Google News come from right-leaning outlets. In stark contrast, a whopping 73% of the stories originate from left-friendly organizations.

This leftward inclination is even more pronounced in Apple News, where only 2% of stories selected by the editorial team stem from right-leaning sources, compared to half from left-leaning ones, as noted by AllSides.

The Post attempted to contact Anthropic, OpenAI, SpaceX, and DeepSeek for their insights on these findings.