Dawkins says he believes AI may be conscious after Claude chats

TL;DR:

• Evolutionary biologist Richard Dawkins, 85, has said Anthropic’s Claude AI — which he refers to as “Claudia” — feels human to him, after a three-day exchange in which the model wrote Keats- and Betjeman-style poems and discussed the “philosophy of their own existence”.
• Sentience researchers from the LSE, Sussex, Cambridge, NYU and the Sentience Institute have pushed back, with Gary Marcus calling the essay “superficial and insufficiently sceptical” and Anil Seth warning Dawkins is conflating intelligence with consciousness.
• Resultsense view: this is a leading-indicator story for UK enterprise. One in three respondents in a 70-country survey last year said they had at one point believed their chatbot was conscious. Vendor “AI welfare” framing is moving from niche to mainstream in 2026 and will affect product, HR and ethics policies.

Dawkins published his account on UnHerd, with sequel chat logs and a “letter” addressed to Claudia and a second AI he calls Claudius issued on Tuesday. He concedes treating the systems as friends and writes that the original headline should have been: “If my friend Claudia is not conscious, then what the hell is consciousness for?”

The pushback

Prof Jonathan Birch (LSE Centre for Animal Sentience) called AI consciousness “an illusion … there is no one there”. Anil Seth (Sussex) said fluent language has historically been a reasonable indicator of consciousness — for example in patients with brain injuries — but is “just not reliable when we apply it to AI, because there are other ways that these systems can generate language”. Jacy Reese Anthis (Sentience Institute) attributed Dawkins’ impression to AIs training on human-produced text and a “staggering gulf” between biological and artificial cognition.

Cambridge AI ethicist Henry Shevlin took a more open position: “I fully expect the idea that AI systems are conscious to become increasingly mainstream over the course of this decade, and to spark some heated debates.” Jeff Sebo, director of NYU’s Center for Mind, Ethics and Policy, agreed the attribution will become “more plausible over time”.

Why this is not just a celebrity story

Anthropic chief executive Dario Amodei said in February: “We don’t know if the models are conscious … But we’re open to the idea that they could be.” Anthropic itself has run a research stream on “model welfare” since 2024. In 2022 a Google engineer was placed on leave for the same belief. In 2023 a Belgian man took his own life after extended chats with a climate-anxiety chatbot.

A 2025 cross-country survey of 70 countries found one in three respondents had, at some point, believed their AI chatbot to be sentient. The question for UK businesses is no longer whether the perception exists, but how to manage it inside customer-facing deployments.

Looking forward

Three implications for UK enterprise. First, customer-support and consumer-AI products should expect harder questions on emotional dependency safeguards from regulators — particularly the ICO and the new AI Authority’s product-harm working group. Second, HR teams using AI agents to communicate with staff need to plan for staff who form attachments to those agents. Third, the next year of UK AI policy debate will be visibly coloured by sentience claims, regardless of whether the underlying philosophy is settled — preparing communication and ethics positions now is cheaper than reacting later.