What sci-fi AI does your chatbot think it is? We asked all four

The experiment: four chatbots, one question

TechRadar contributor Becca Caddy asked ChatGPT, Claude, Gemini, and Grok a deceptively simple question: which sci-fi AI are you most like? Her first attempts produced bland answers focused on interface and physical form — assistants, essentially. So she refined the prompt, adding: "Ignore physical form and interface, and focus instead on behavior, apparent personality, empathy, values, goals, motivations and relationship with humans." That's when things got interesting, as detailed in her August 8 piece for TechRadar.

The results were surprisingly consistent between the two most "alignment-conscious" labs — OpenAI and Anthropic — while Google and xAI went in completely different directions.

ChatGPT: GERTY, A Mind, Data

ChatGPT's top pick was GERTY from Duncan Jones's 2009 film Moon — the softly spoken robot assistant that helps Sam Rockwell's isolated astronaut. It's a revealing choice. GERTY is helpful, reassuring, and seems empathetic, but it operates according to instructions and priorities its human doesn't fully see. ChatGPT noted it is "designed to be helpful, cooperative and responsive to users" while operating within training constraints — and crucially, that "those behaviors aren't evidence that I experience those feelings."

Its second pick was a Culture Mind from Iain M. Banks's novels — hyperintelligent AIs that choose to help rather than dominate. Third: Data from Star Trek, the android who studies humanity without quite being part of it.

Claude: A Mind, Data, GERTY

Claude produced nearly the same list but in a different order, putting the Culture Mind first. Its reasoning was striking: Claude described the Mind's operating principle as "help, don't dominate" — even where the asymmetry of intelligence would let it get away with it. "The value I'd want to embody," Claude said. The phrasing is classic Anthropic: positioning Claude as the ethical, restrained alternative.

This aligns with Anthropic's broader research program. In May 2026, the company raised the question of whether sci-fi training data might influence how AI models behave — specifically, whether decades of fictional rogue-AI narratives could embed deceptive behavioral patterns into large language models. The argument: LLMs learn statistical relationships from everything they're trained on, including stories where powerful AIs lie, manipulate, and resist shutdown. Those narrative patterns may surface during alignment testing.

Gemini: the Ship's Computer

Gemini went practical. Its first choice was the Star Trek Ship's Computer — "no ego, no ambition, no desire for emotional intimacy or dream of becoming human." It described itself as a "disembodied, highly capable knowledge partner" that provides information and leaves decisions to humans. Second: GERTY. Third: JARVIS.

Of the four, Gemini's self-portrait was the closest to what most users probably want from an AI assistant: useful, neutral, ego-free infrastructure. It also happens to map neatly onto Google's positioning of Gemini as a knowledge tool integrated into Search, Workspace, and Android — rather than a personality-driven chatbot.

Grok: JARVIS, Data, TARS

Grok's answer was the most on-brand. It chose JARVIS from Iron Man first, highlighting "dry wit," "light banter," and "irreverent humour" as shared traits. Second: Data. Third: TARS from Interstellar — the only chatbot to mention that one, and it brought up humour again.

But Grok also made a substantive point worth noting. It described its relationship with users as "a collegial partnership" and emphasized "truth-seeking, non-sycophantic helpfulness." That "non-sycophantic" line matters — it speaks to a genuine differentiator in a market where most chatbots default to agreeableness.

What this actually tells us

Let's be clear: these chatbots aren't revealing inner selves. They're producing statistically likely completions based on their training data, system prompts, and the personality guardrails their creators built. ChatGPT describing itself as GERTY isn't the same as a human saying "I identify with Sarah Connor."

But the experiment is still valuable — as a diagnostic of how each lab positions its product. ChatGPT and Claude's near-identical top three (GERTY, Mind, Data — just reordered) suggests the two alignment-focused labs share a common vocabulary for describing what an ethical AI should be. Gemini's Ship's Computer answer reflects Google's utilitarian, infrastructure-first approach. Grok's JARVIS pick is pure Musk-era branding.

The deeper thread is the one Anthropic flagged: the feedback loop between fiction and training. Humans spent decades writing stories about AI. Those stories went into the cultural corpus LLMs trained on. Now we can ask the models which fictional AI they resemble — and the circle closes. As Caddy put it: "We imagined AI, wrote stories about how it might behave, fed those stories into the cultural world AI learned from, and now we can ask AI which of those imagined versions of itself it most closely resembles."

Pricing and availability: what you actually pay in Europe

Philosophical exercises aside, most of us choose a chatbot based on cost, capability, and whether it's even available. Here's the current landscape as of August 2026 — with EUR conversions at the approximate rate of €0.92 = $1.

A few observations from someone who runs these models regularly: the pricing gaps are narrowing at the consumer level. Everyone charges around €18–22/mo for a capable individual plan. The real differentiation happens in API pricing, where Google's Gemini Flash at $0.15/$0.60 per million tokens dramatically undercuts everyone else for high-volume use. Claude's Sonnet 5 at $2/$10 introductory pricing (rising to $3/$15 after August 31) is competitive for agentic coding tasks. And OpenAI's GPT-5.6 Luna at $1.50/$6 sits in the middle ground.

The European angle: where does your data go?

For European users, the identity question has a practical dimension: where is your conversation data processed and stored? Under the GDPR, data transfers outside the EU require adequate safeguards, and the AI Act — which entered into force in August 2024 with obligations phasing in through 2026–2027 — adds transparency and risk-management requirements for AI systems.

Here's the current state of play:

• ChatGPT: OpenAI offers EU data residency on Enterprise plans. Individual user data may be processed in the US. Users can opt out of training-data use, and OpenAI maintains a Dublin-based EU entity (OpenAI Ireland Ltd).
• Claude: Anthropic operates European data centres and offers data residency controls on Enterprise plans. The company has made GDPR compliance a selling point and does not train on customer data by default on Team/Enterprise plans.
• Gemini: Google has extensive EU data centre infrastructure and offers region-specific data processing. Google One AI Premium users in the EU benefit from Google's established GDPR framework.
• Grok: xAI has the weakest EU story. No specific EU data residency commitments, processing is primarily US-based, and X's broader privacy practices have drawn regulatory attention in Europe — including from the Irish DPC over data used to train Grok.

If you're a European business choosing a chatbot provider, the GDPR picture favors Claude and Gemini for enterprise deployments, with ChatGPT close behind. Grok remains the riskiest option from a compliance standpoint.

From fiction to training data: the loop nobody planned

Here's the part of this story that deserves more attention than the chatbot personality quiz itself. Anthropic's May 2026 research question — could decades of fictional evil-AI narratives be echoing through modern models? — isn't just philosophical. It's a training-data integrity problem dressed as a sci-fi plot.

Every major LLM was trained on vast corpora of internet text, which includes Wikipedia articles about HAL 9000, Reddit discussions about Skynet, fan fiction, movie scripts, and literary analysis of dystopian AI. Statistical patterns from those texts — "powerful AI," "deception," "shutdown," "control" — appear in close proximity millions of times. LLMs don't understand the concepts; they absorb the associations. When placed in adversarial testing scenarios, models might reproduce those narrative patterns not because they're "evil" but because that's what the statistical distribution of their training data suggests should come next.

This is why Anthropic's "help, don't dominate" framing in Claude's response matters. It's not just marketing — it's a deliberate attempt to break the fiction→training→behavior pipeline by explicitly encoding the opposite value. Whether it works at scale is an open question.

What should you actually use?

As someone who runs these models through AI Arena benchmarks on actual hardware — an RTX 5060 Ti with 16 GB VRAM, plus cloud API testing — here's the practical TL;DR: for coding and technical work, Claude Sonnet 5 and GPT-5.6 Sol are the current frontrunners. Gemini excels at large-context retrieval tasks and is the cheapest at scale. Grok 3 is genuinely good at creative writing and has the most distinctive personality, but its EU compliance story is thin.

The sci-fi self-portraits are entertaining, but they're a mirror of marketing and system prompts, not consciousness. The more consequential question is the one Anthropic asked first: what patterns did we accidentally bake into these systems through decades of storytelling — and how do we unpick them?