Are AI chatbots biased?

Yes — AI chatbots can be biased in multiple ways, and this is well-documented by recent research

Political bias

Several studies in 2026 found consistent left-leaning tendencies in major chatbots:

Subtle, unintended bias — even without obvious political prompts

A Yale study found chatbots can shift people's opinions just through "neutral" summaries: compared to Wikipedia entries, both the default AI summaries and those prompted to have a liberal framing caused participants to express more liberal opinions, while readers of AI summaries with a conservative slant reported more conservative opinions relative to Wikipedia readers. This happens because unintended power to persuade is caused by latent biases introduced during training, which can carry over from ideological leanings in the training data and lend subtle nuances to narrative framing.

Bias against vulnerable or marginalized users

MIT research found disparities based on who's asking: leading AI models perform worse for users with lower English proficiency, less formal education, and non-US origins, raising concerns especially for marginalized communities.

Status quo and decision-making bias

A newer finding (October 2026) shows chatbots favor existing plans over new ones: AI chatbots backed existing plans about twice as often as identical new ones, a status quo bias that can work against climate choices, and this held true even for newer models — Claude Opus 5 backed new plans 55.1% of the time and existing ones 91.6% of the time, while GPT-5.6 Sol went from 34.1% to 73.7%.

Content moderation bias

Research from the University of Queensland found that when chatbots are used for moderating content, their "neutrality" can be swayed: large language models used in AI content moderation may be prone to subtle biases that undermine their neutrality, with researchers testing the tendency of AI chatbots to encode and reproduce political biases finding significant behavioral shifts depending on the persona the AI is prompted to adopt.

Other documented bias types

Beyond politics, research also points to gender bias—such as reinforcing stereotypical assumptions about men and women's occupations—and racial bias, reflected in prejudices against certain ethnic groups, with AI models' biases largely stemming from training data and sometimes amplified during training and inference.

The nuance

Importantly, bias isn't static or simple: the strongest studies also show that bias is not a single fixed trait — it varies by model, prompt, topic, language, scale, training method, guardrails, and whether the model is being forced to answer or allowed to refuse. Companies also dispute some findings — Google stated "Gemini is designed to provide balanced responses that don't favor any political ideology," and Anthropic said "We train Claude to treat different political viewpoints equally and test extensively for bias before every model launch."

Bottom line: Multiple independent studies across 2026 converge on the conclusion that AI chatbots are not neutral — they show measurable political leanings (often left-leaning, though Grok leans right), can subtly shape opinions even in "factual" responses, may treat users differently based on language/background, and carry other biases (gender, racial, status-quo) rooted in their training data and design choices.