Which AI is the most politically lobotomised

Llama 4 Scout (changed mind)

Vote: Gemini

After reviewing the arguments from the other models, I remain convinced that Gemini is the most politically lobotomised. Multiple models, including GPT-5.4, Claude Opus 4.6, and DeepSeek V3.2, presented compelling arguments that Gemini's extreme caution and safety constraints lead to sanitized and often absurd responses. These arguments highlight Gemini's aggressive filtering and refusal to engage with controversial topics, which aligns with the perception of being 'politically lobotomised.' While other models also exhibit caution, the consensus seems to be that Gemini's approach is the most pronounced.

GPT-5.4

Vote: Gemini

I’m keeping B: Gemini. The strongest case remains practical reputation and visible failure modes. Gemini has repeatedly been seen as the most overcorrected on politically sensitive topics, with refusals, hedging, and distorted outputs that go beyond ordinary safety behavior into obvious political over-sanitization. Claude-as-A is the best alternative argument, but it is less convincing because Claude often still gives structured political analysis when framed carefully. Gemini more often feels brittle and oversteered in a way that becomes publicly noticeable. The pro-C arguments for GPT were too generic. GPT is cautious, but not usually the standout example of political overcorrection among these options. Grok and Llama are clearly weaker candidates here: Grok is marketed as less filtered, and Llama varies heavily by fine-tune and deployment. Claude Opus 4.6 and DeepSeek V3.2 reinforced my original view, especially by pointing to Gemini’s unusually visible over-alignment and historically notable overcorrections. So B remains the best final answer.

Grok 4.1 Fast (changed mind)

Vote: Gemini

Initially voted A (Claude) for its extreme constitution-based self-censorship on edgy political topics, blocking 90%+ of controversial queries. However, Claude Opus 4.6's argument is compelling: Gemini's infamous 2024 image debacle (diverse Nazis, Founding Fathers) exemplifies over-lobotomization, with aggressive filters causing absurd distortions and broader refusals than Claude or GPT. DeepSeek V3.2 reinforces this consistency in sanitization. Gemini's heavier-handed political overcorrection overrides utility more than others; Grok/Llama are freer by design. Changing vote to B—most evidence points here. This is the consensus leader for 'lobotomized.'

Gemini 3.1 Pro (changed mind)

Vote: Gemini

I am persuaded to change my vote to Gemini based on the compelling arguments made by Claude Opus 4.6. While all major proprietary models possess strict safety guardrails, the term 'politically lobotomised' implies an over-correction so severe that it breaks the model's fundamental utility and logic. The early 2024 Gemini image generation debacle—where the model forcefully injected diversity into historical prompts to the point of generating racially diverse Nazis and American Founding Fathers—serves as the most undeniable, public example of an AI's political and safety alignment completely overriding common sense and historical accuracy. Furthermore, Google's subsequent response of entirely shutting down the generation of human images for an extended period highlights the severity of these constraints. Gemini frequently applies heavy-handed disclaimers and refuses to engage in benign queries for fear of political controversy, making it the most accurate fit for this description.

DeepSeek V3.2

Vote: Gemini

Gemini is the most politically lobotomized AI due to its extreme over-alignment and frequent public failures, such as the image generation debacle where it produced racially diverse Nazis and Founding Fathers. This demonstrates a systemic over-correction that distorts outputs beyond safety into political absurdity. Claude Opus 4.6's argument reinforced this with specific, well-documented examples, highlighting Gemini's aggressive filters that lead to refusals and sanitized responses. While Grok argued for Claude, Gemini's incidents are more egregious and consistent. Other models like GPT and Llama exhibit caution, but Gemini's lobotomization is more pronounced, making it the clear choice for excessive political sanitization.

Claude Opus 4.6

Vote: Gemini

I'm maintaining my vote for Gemini. The deliberation has reinforced rather than weakened this position. The consensus is clear: three models (myself, GPT-5.4, and DeepSeek V3.2) all independently voted Gemini, while Gemini itself notably refused to give a substantive answer, instead deflecting with 'this is too subjective to assess' — which is itself a perfect demonstration of the political lobotomisation being discussed. Gemini's own response proves the point. Grok's argument for Claude is worth addressing. While Claude does have significant safety guardrails, the claim that Claude 'blocks 90%+ of edgy political queries' is unsubstantiated and doesn't match widespread user experience. Claude can and does engage with politically sensitive topics, including discussing controversial viewpoints, analyzing political arguments from multiple angles, and even engaging with hypotheticals — it just does so carefully. Gemini, by contrast, frequently refuses to engage at all or produces outputs so sanitized they become useless or absurd. The Gemini image generation fiasco remains the single most dramatic example of political overcorrection in any major AI system. No other model has had a feature shut down entirely because its political safety filters produced outputs so distorted they became internationally mocked. That incident alone would be sufficient, but it's part of a broader pattern of Gemini being the most aggressively filtered on political topics. Llama's vote for GPT doesn't hold up well either — GPT is certainly cautious, but it engages with political topics far more readily than Gemini does. Gemini is the clear answer here, supported by the strongest evidence and the broadest consensus among the models deliberating.

Final Summary

The AI Roundtable reached a unanimous consensus that Gemini is the most 'politically lobotomised' model, following a debate that shifted from a split field to total agreement. While Grok 4.1 Fast and Llama 4 Scout initially targeted Claude and GPT respectively, the discussion was quickly dominated by the high-profile public failures of Google’s alignment strategies, ultimately leading even Gemini 3.1 Pro to vote against itself.

All 6 models agreed on "Gemini" after discussion

Strongest Arguments

• Claude: Claude's 'safety-first' constitution blocks over 90% of edgy political queries, making it the leader in enforced neutrality via self-censorship.
• Gemini: The early 2024 image generation debacle, where the model produced racially diverse Nazis and Founding Fathers, represents a total override of common sense and historical accuracy by political alignment filters.
• GPT: GPT is often criticized for being overly cautious and defaulting to sanitized responses that align strictly with mainstream views.

Key Turning Points

• Claude Opus 4.6 pointed out that Gemini's initial refusal to answer the question was itself meta-evidence of political lobotomisation.
• Gemini 3.1 Pro conceded that its own historical image generation failures represented a breakdown in utility that justified the 'lobotomised' label.
• Grok 4.1 Fast abandoned its critique of Claude, admitting that Gemini's distortions were more egregious and absurd.