Margaret Atwood used Anthropic's Claude exactly once, asked it about a British detective show, and walked away convinced the technology is overhyped. The author of The Handmaid's Tale and The Blind Assassin made the comments at the Babell Literary and Cultural Festival in Porto, Portugal, on June 27, 2026, summarising the entire category in five words: garbage in, garbage out.
Atwood's test case was specific. She asked Claude about the long-running ITV series Father Brown — and Claude, she said, returned an answer that was either wrong or invented. The model had clearly read a great deal about the show. It had not actually watched it, and online reviews of detective fiction tend to omit the one fact a viewer might want from a summary: who did it.
That detail matters because it gets at a genuine structural limit of large language models rather than a bug in any single product. Claude is trained on text scraped from the open web. If the open web declines to spoil the ending of a cozy mystery, Claude has no clean signal to retrieve, and the model defaults to plausible-sounding synthesis. Atwood, an 86-year-old novelist whose entire career is built on plot, noticed immediately.
“Claude gave me the wrong answer, or it lied. Of course, it didn't know it was lying because it's not a human being; it's a large language model... It had skimmed and sampled a lot of television reviews, but they never give away the ending in online criticism, so it was misled by the things it had read about the show.”— Margaret Atwood, Author of The Handmaid's Tale
Key facts
- 01Margaret Atwood said she used Anthropic's Claude exactly once, to ask about the British detective series Father Brown, and got a wrong answer.
- 02Atwood spoke at the Babell Literary and Cultural Festival in Porto, Portugal, on June 27, 2026.
- 03She argued LLMs are constrained by training data: reviews of Father Brown rarely spoil the ending, so Claude guessed incorrectly.
- 04Atwood called AI users 'opportunists' and warned business users must check outputs because the model 'makes mistakes.'
- 05The author of The Handmaid's Tale and The Blind Assassin has not endorsed using AI in her writing process.
Her broader critique was directed at users rather than vendors. Atwood described people who lean on AI as opportunists looking for the path of least resistance, and she made the point that even commercial users have to manually verify what the model produces. That framing — AI as a productivity tool whose output is provisional until checked — is closer to how careful enterprise buyers actually deploy these systems than the marketing implies.
Anthropic has not responded to Atwood's remarks, and there is no indication the company sees a novelist's one-shot test as a product issue. Claude's known weaknesses on niche cultural trivia and long-tail entertainment metadata are well-documented inside the AI evaluation community; the model is tuned far more aggressively for coding, analysis, and long-context reasoning than for trivia recall about regional British television.
Atwood's comments arrive at an awkward moment for the publishing world, which is in active litigation with several model providers over training data. Authors have sued OpenAI, Anthropic, and Meta over the use of copyrighted books in pretraining corpora, and the cases continue to work through US courts. Atwood did not raise the copyright question in Porto, but her framing — that the models are derivative by construction — is exactly the argument plaintiffs in those cases are making.
“Human beings are not robots, but they are opportunists, so if there's an easy way to cheat and it's hard to detect, people will do it... But the thing about AI is that it's garbage in, garbage out. Even people who use it for business reasons have to check it because it makes mistakes.”— Margaret Atwood, Author of The Handmaid's Tale
There is also a generational read on this. Atwood is part of a cohort of literary figures, including Nobel laureates and major novelists, who have been broadly skeptical of generative AI's encroachment on creative work. The skepticism tends to land harder on factual reliability than on aesthetics — writers can usually tell when prose is machine-generated, and they are unimpressed by what they read. Atwood's choice of phrase, recycled from decades-old computer science folklore, was deliberate.
What Atwood did not say is also worth noting. She did not call for AI to be banned, did not demand new regulation, and did not suggest writers boycott the tools. Her objection is that the tools do not work well enough to be trusted, which is a quality complaint rather than a moral one. That is a useful distinction at a moment when public AI commentary tends to collapse those two categories.
The interesting business question underneath Atwood's anecdote is whether labs care about the long tail. Claude, GPT-5.6, and Gemini are all increasingly optimized for agentic coding, research workflows, and high-value enterprise tasks where the buyer is paying hundreds of dollars per seat per month. Trivia accuracy on a 1980s detective adaptation is not on anyone's roadmap. The risk for Anthropic is reputational drag — every Atwood-style anecdote that goes viral chips away at the case that these systems are ready for general consumer trust, even if the enterprise contracts keep closing.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
