Anthropic thinks its AI might be conscious

Artificial intelligence can reason, solve problems, and discuss its own apparent feelings. But does any of this mean that a machine is conscious? Recent research involving Anthropic’s Claude models and the company’s latest statements on AI consciousness has brought this longstanding philosophical question to the center of the AI debate.

In July 2026, Anthropic researchers described an internal workspace in Claude, called J-space, that appears to help the model represent and manipulate concepts independently of its immediate output. The research has renewed speculation about whether advanced AI might possess something resembling consciousness. However, identifying internal representations is not proof of subjective experience. Anthropic’s findings likewise distinguish limited introspective capabilities from evidence that a model is conscious.

This is a distinction that matters. A system can process information about its internal states without necessarily experiencing those states. In humans, consciousness involves subjective experiences, known as qualia: what it feels like to see a sunset, taste coffee, or feel pain and emotion. Whether LLM computation alone can produce such experiences remains unresolved.

What does science say about conscious AI?

Researchers disagree about what consciousness requires and how it could be detected in a non-biological system. Some theories suggest that particular forms of information processing could, in principle, support consciousness in machines. Others argue that intelligence and convincing human-like behavior are insufficient.

A significant contribution to this debate was published in Trends in Cognitive Sciences in June 2026. The paper, Identifying indicators of consciousness in AI systems, proposes evaluating AI using indicators derived from established neuroscientific theories of consciousness. Rather than relying on a chatbot’s claims about what it's feeling, the researchers argue for a more systematic assessment.

A challenge to this view came from neuroscientists Giulio Tononi and Anil Seth and the philosopher Ned Block. In an October 2026 commentary for Nature, Three reasons why AI is not conscious and perhaps never will be, they argue that currently existing AI systems are not conscious and that intelligence alone does not establish subjective experience.

This position is influential, but it does not settle the debate. The responsible conclusion is that we have evidence of increasingly sophisticated information processing, but not established evidence of machine consciousness.

If we look into what some of the greatest minds in human history have said about the possibility of machine consciousness, what guidance do they have to offer us?

Stephen Hawking’s warning becomes relevant

The famous physicist Stephen Hawking approached AI from another angle: he warned about the potential consequences of creating intelligence that could surpass human capabilities. In a 2014 interview with the BBC, Professor Hawking warned that “the development of full artificial intelligence could spell the end of the human race.” He was concerned that a sufficiently advanced system might improve itself faster than humans could adapt.

“The development of full artificial intelligence could spell the end of the human race.”
—Professor Stephen Hawking

His warning was about control of vital systems, not proof that AI would automatically become conscious. A machine could potentially pose serious risks through its capabilities and objectives even without having feelings or subjective awareness.

Can machines think?

British mathematician, logician, cryptographer, and computer scientist Alan Turing did not offer a complete theory of phenomenal consciousness. Nor did he claim that passing his famous Turing Test proved a machine was conscious. Instead, his view was pragmatic, behavioral, and methodologically agnostic. In his 1950 paper Computing Machinery and Intelligence, Turing famously asked, “Can machines think?”

Rather than getting bogged down in unanswerable metaphysical debates about what inner experience or a soul really is, he replaced it with an operational, observable challenge known as ‘the imitation game.’ For Turing, intelligence could be learned.

“We may hope that machines will eventually compete with men in all purely intellectual fields.”
—Dr. Alan Turing

Nicola Tesla’s remarkable vision

Long before modern computing, inventor Nikola Tesla was exploring the relationship between intelligence, automation, and remote control. In 1898, Tesla demonstrated a remotely controlled exhibition in New York. His invention, which he called a ‘Teleautomaton,’ was a type of uncrewed boat that responded to wireless control signals and incorporated propulsion, steering, and other mechanisms. Tesla described it as a more than conventional mechanical device, envisioning machines capable of performing tasks with something resembling intelligence and judgement.

The Teleautomaton can reasonably be considered an ancestor of modern robotics, unmanned vehicle technology, and wireless remote control. It was not an AI robot in the contemporary sense; its behavior was governed by engineered control mechanisms, not a modern learning system. Nevertheless, Tesla demonstrated that machines could act at a distance without a human physically directing every movement. In his June 1900 essay The Problem of Increasing Human Energy, Tesla went even further. He imagined automata equipped with sensing, movement, and mechanisms for responding to external stimuli, capable of performing tasks with apparent judgement. Tesla anticipated important principles of autonomous machinery, although not artificial consciousness itself.

The question beyond intelligence

More than a century after Tesla’s experiments, humans are building machines that can perform tasks once considered uniquely human. Yet the ability to imitate reasoning is not necessarily the ability to experience it. Perhaps the most important question to ponder is not whether AI can convince us that it thinks, but how we could ever determine what it is like to be an AI system.

Until researchers can answer that question with convincing evidence, we should neither dismiss the possibility outright, nor mistake increasingly human-like behavior for proof of a conscious mind. Or, until a machine system convincingly proclaims with Cartesian certainty, “Cogito, ergo sum”—“I think, therefore I am.”