Margaret Atwood has never been shy about uncomfortable futures. This week, the author of The Handmaid’s Tale and The Blind Assassin turned her attention to artificial intelligence, and her verdict was blunt: AI is only as reliable as the material it has been fed.
Speaking at the Babell Literary and Cultural Festival in Porto, Portugal, Atwood said she had tried an AI chatbot exactly once. The chatbot was Anthropic’s Claude, and the experience did not make her a convert.
Margaret Atwood on AI: Why ‘garbage in, garbage out’ still matters
According to Deadline’s recap of the event, Atwood said she asked Claude for information about the British detective series Father Brown. Instead of delivering a useful answer, she said the chatbot gave her one that was simply wrong.
Atwood described the exchange sharply, saying Claude “gave me the wrong answer, or it lied.” She immediately clarified the point: a chatbot is not a person, so it does not understand deception in the human sense. It is a large language model generating text based on patterns in its training data and prompts.
That distinction sits at the center of her criticism. AI tools can sound polished, certain, and helpful even when they are off base. For writers, researchers, students, and anyone using AI search tools, that confidence can be the real problem.
Claude, hallucinations, and the risk of confident wrong answers
Atwood’s story is a familiar one for anyone who has tested AI chatbots beyond simple tasks. Large language models can summarize quickly, draft fluently, and connect ideas at impressive speed. They can also invent details, misread sources, or present shaky information as fact.
In AI circles, these mistakes are often called “hallucinations.” The term may sound almost whimsical, but the consequences are not. A wrong answer about a TV programme is irritating. A wrong answer about health, law, finance, history, or authorship can be far more serious.
Atwood’s “garbage in, garbage out” point is not new in computing, but it has fresh urgency in the chatbot era. If a system learns from messy, biased, incomplete, or low-quality material, its output can reflect those problems. The smoother the writing, the easier it is to miss the flaw.
Why authors are watching artificial intelligence so closely
Atwood’s comments also land in a wider debate about AI and creative work. Authors, journalists, artists, and publishers have spent the past few years asking hard questions about how AI models are trained, whether copyrighted work has been used without permission, and how machine-generated text should be labeled.
For a novelist like Atwood, the concern is not only whether AI can write a passable paragraph. It is whether readers, companies, and institutions will treat generated language as trustworthy without checking its foundations.
That is where her criticism cuts deepest. The problem is not just that AI can be wrong. It is that it can be wrong beautifully.
What Atwood’s AI warning means for everyday users
The practical takeaway is simple: use AI as a starting point, not a final source. If a chatbot gives you a fact, trace it back to a reliable reference. If it summarizes a book, show, article, or study, compare the answer with the original. If it sounds too neat, be extra suspicious.
Atwood’s single Claude experiment may be anecdotal, but it captures a larger truth about generative AI in 2026. These tools are powerful, useful, and increasingly embedded in daily life. They are also fallible systems that can mistake probability for knowledge.
For readers curious about Father Brown, the series is a BBC detective drama. It is available in the UK through BBC iPlayer, while US availability commonly includes BritBox; access across the EU can vary by country and licensing window.
Atwood has spent much of her career warning audiences about systems people trust too easily. Her latest AI criticism fits that tradition perfectly: before believing the answer, look closely at what went in.
Tags: #MargaretAtwood #ArtificialIntelligence #ClaudeAI #AIEthics #GenerativeAI
About the Author
