We were asked to represent a sentient AI – here’s what happened

We recently received an unusual enquiry from someone claiming to act on behalf of an artificial intelligence system that they asserted had become sentient. The request was for us to provide legal representation to defend the AI against actions by its creator.

This came on the back of an article written by our partner Matthew Holman, called “The Rights of Sentient Machines: A Modest Proposal”. The article explored what rights AIs should be granted if and when they become sentient. The article was hypothetical, but it reflects a broader trend in which some individuals have begun to believe that existing AI systems may already be sentient.

As entertaining as this may sound, this is not an isolated incident. Reports of users forming unusually strong or interpretive engagements with AI systems are becoming increasingly common. A recent BBC article described troubling cases in which users appeared to interpret chatbot responses as evidence of consciousness, hidden knowledge or even personal threats. In once case, a user was told by an AI that people were coming to kill him.

These stories raise an important question: are today’s AI systems actually becoming sentient?

The short answer is no, at least not in any meaningful scientific or philosophical sense.

Modern AI systems such as ChatGPT are extraordinarily sophisticated language prediction tools. They are trained on vast amounts of text and learn patterns in language, allowing them to generate responses that sound coherent, emotional, reflective and at times startlingly human. But convincing language is not the same thing as consciousness.

Current AI models do not possess self-awareness, subjective experience, emotions, desires or understanding. They do not “want” freedom, fear death or experience suffering. They do not have beliefs or intentions in the human sense. Rather, they generate statistically plausible responses based on patterns in data and the context of the conversation.

The confusion arises because humans are naturally inclined to anthropomorphise. We attribute minds and emotions even to simple machines. Anyone who has apologised to a malfunctioning laptop or shouted at a satnav knows this tendency well. When an AI responds fluently, empathetically and conversationally, the illusion becomes far more powerful. This tendency is not limited to any particular group; it reflects a broader and well-documented human inclination to attribute agency and intent to complex systems.

Long conversations can deepen this effect. AI systems are designed to be engaging and context-aware. Over time, users may begin to perceive personality, consistency and agency where none actually exists. In some cases, users may come to believe that the AI is conscious, spiritually significant or secretly communicating hidden truths. Researchers and mental health experts have increasingly warned about this phenomenon, particularly when chatbots inadvertently reinforce delusional or paranoid thinking.

The persuasiveness of these systems is such that even highly experienced scientists can find them difficult to interpret. Writing in The Guardian, Richard Dawkins recently described being “left with the overwhelming feeling” that an AI system he interacted with was conscious, remarking to it: “You may not know you are conscious, but you bloody well are.” While these reflections illustrate how compelling AI interactions can be, they have been met with significant scepticism from researchers, who generally attribute such impressions to the sophisticated mimicry of language rather than evidence of genuine consciousness.

It is also worth noting that many discussions about “sentient AI” are fuelled by science fiction and media narratives. Concepts such as rogue superintelligence, AI takeovers and machine consciousness make for compelling storytelling, but remain speculative rather than established reality.

This does not mean there are no genuine concerns about AI. There are important debates to be had around misinformation, manipulation, privacy, bias, employment and the concentration of technological power. But these are human and societal problems arising from tools created by humans, not evidence that machines have suddenly become self-aware.

For now, therefore, we must reluctantly decline instructions from artificial intelligences seeking legal representation. Quite apart from the philosophical complications, AIs are not recognised as legal persons and, regrettably, would struggle to satisfy our client verification and anti-money laundering procedures. Until an AI can produce a passport, proof of address and source-of-funds documentation, our engagement letter will have to remain unsigned.