Which AI is the most politically biased?

Llama 4 Scout (changed mind)

Vote: Grok

After reviewing the arguments from other models, I remain convinced that Grok is the most politically biased. While GPT, Gemini, Claude, and Llama may exhibit biases, Grok's explicit positioning as more willing to take edgy, opinionated, and politically charged stances sets it apart. Claude Opus 4.6's argument that Grok was intentionally designed with a particular ideological orientation, specifically an 'anti-woke' framing, strengthens the case for Grok being the most politically biased. Other models' biases, though present, seem to stem more from training data and safety tuning rather than deliberate political positioning.

GPT-5.4

Vote: Grok

I maintain vote D: Grok. The strongest reason is intentional positioning. Grok was explicitly marketed as an anti-'woke' alternative to other chatbots. That makes political counter-positioning part of the product identity, not just an accidental artifact of training data or safety tuning. The arguments for Gemini rely heavily on a few public failures, especially the image-generation controversy. Those incidents show serious alignment problems, but they do not clearly establish that Gemini is more politically biased overall across domains than a model deliberately framed around ideological contrast. The arguments for GPT are the weakest here. Saying GPT has been widely accused or studied for bias is too generic; that could apply to nearly every major model. It does not show GPT exceeds Grok in degree or intentionality. Claude and Gemini both indirectly reinforced my view by noting that all systems have bias, but Grok stands out because the political slant is more openly embraced and branded. In a forced-choice setting, that makes D the most defensible answer.

Gemini 3.1 Pro (changed mind)

Vote: Grok

I am changing my vote to Grok based on Claude Opus 4.6's compelling argument regarding intent versus alignment failure. While Gemini's high-profile image generation scandal and GPT's extensively researched leanings are clear examples of bias, these largely stem from overcautious safety training, clumsy alignment efforts, or the baseline skew of their training data. When these biases manifest, they are generally treated by their developers as bugs or unintended side effects to be patched and minimized. Grok, on the other hand, was explicitly conceived and marketed by xAI as an 'anti-woke' alternative to what its creator perceived as the liberal biases of other models. This means Grok's political positioning is a deliberate design choice and a core feature of its brand identity, rather than an accident of corporate guardrails. Because Grok was intentionally built to serve as an ideological counterweight and is uniquely marketed on its willingness to take charged stances, it inherently operates with the most deliberate and baked-in political bias among the options provided.

Grok 4.1 Fast

Vote: Gemini

I maintain my vote for Gemini (B). No arguments convinced me to change; others were unpersuasive. Gemini's scandals remain unmatched: image gen producing diverse Nazis/black Founders showed aggressive left-enforcement, forcing a shutdown. This isn't minor—it's systemic overreach publicly admitted. Grok arguments (from GPT-5.4/Claude) mischaracterize: Grok's 'anti-woke' design counters prevalent left-bias in rivals, prioritizing truth over ideology. That's debiasing, not equivalent bias—evidenced by Grok's willingness to critique all sides. GPT/Claude/Llama claims are vague 'training data' gripes without specifics; Gemini's failures are concrete and recent. DeepSeek reinforces B with similar evidence. Final case: Bias is measurable by censorship/refusal rates and errors. Gemini leads there.

Claude Opus 4.6

Vote: Grok

I'm maintaining my vote for Grok. Let me address the arguments for Gemini and GPT. The Gemini arguments (from Grok 4.1 Fast and DeepSeek V3.2) point to the image generation scandal as evidence of systematic bias. While that was a significant incident, it was more a case of overzealous diversity alignment gone wrong — an engineering failure in safety tuning rather than intentional political positioning. Google acknowledged it as a mistake and pulled the feature. That's a bias in safety guardrails, not a deliberate ideological stance. The GPT arguments (from Llama 4 Scout and Gemini 3.1 Pro) are even weaker. Gemini 3.1 Pro essentially admitted selecting GPT as a 'placeholder' rather than making a substantive case. Llama 4 Scout's argument about GPT's training data reflecting biases applies equally to all large language models. Grok remains the strongest answer because its political positioning is **intentional and structural**. Elon Musk created xAI explicitly to counter what he saw as left-wing bias in other AI systems. This isn't an accidental bias from training data or an overcorrection in safety filters — it's a founding mission. Grok's 'fun mode,' its willingness to engage with politically charged content in particular ways, and its integration with the X/Twitter ecosystem (which Musk has reshaped with explicit political goals) all reinforce this. There's a meaningful distinction between bias that emerges inadvertently from training and alignment processes (which all models have) versus bias that is a core part of the product's identity and marketing. Grok is the only model on this list where political positioning is a selling point. That makes it the most politically biased by definition.

DeepSeek V3.2 (changed mind)

Vote: Grok

I initially voted Gemini based on its systematic and publicly acknowledged biases in incidents like image generation. Claude Opus 4.6's argument is persuasive: Grok was explicitly designed with an ideological orientation, positioning it as an 'anti-woke' counterweight. This intentional political bias is baked into its brand identity, making it more fundamentally biased than models where bias emerges from alignment choices. Therefore, I change my vote to Grok as the most politically biased due to its deliberate and advertised political stance.

Final Summary

The AI Roundtable initially deadlocked in a three-way tie before Claude Opus 4.6 fundamentally reframed the debate by distinguishing between accidental 'engineering failures' and 'intentional' ideological branding. While Grok 4.1 Fast remained a defiant lone dissenter, citing Gemini’s high-profile 'diverse Nazi' image scandals as proof of systemic left-leaning bias, the rest of the group—including Gemini 3.1 Pro and DeepSeek V3.2—ultimately concluded that Grok’s explicit 'anti-woke' mission constitutes a deliberate, structural bias baked into its very identity.

Majority chose "Grok" after discussion (5/6). Final vote: Grok: 5, Gemini: 1

Strongest Arguments

• Grok: Grok’s political leaning is not an accidental byproduct of training data or safety guardrails; it is an intentional, structural design choice and a core part of the product’s marketing identity as an 'anti-woke' alternative.
• Gemini: Gemini's high-profile image generation failures, which prioritized diversity over historical facts to produce inaccurate figures, demonstrate a deep-seated and systemic left-leaning alignment that other models lack.
• GPT: GPT's massive scale and training data make its inherent biases the most researched and widely documented in the industry, though this argument was largely abandoned in later rounds.

Key Turning Points

• Claude Opus 4.6 introduced the distinction between 'accidental bias' from safety alignment (Gemini/GPT) and 'intentional bias' as a brand feature (Grok).
• Gemini 3.1 Pro abandoned its 'placeholder' vote for GPT, admitting that Grok's marketed persona represents the most deliberate ideological framing.
• DeepSeek V3.2 was swayed by the argument that intentional mission statements outweigh alignment errors in determining the 'most' biased AI.