Is AI conscious? What the researchers actually say

The short answer: nobody knows, and the experts say so

The most honest summary comes from a company that builds these systems. When Anthropic launched a research program on the question in April 2025, it wrote:

“There's no scientific consensus on whether current or future AI systems could be conscious”

That remains true as of September 2026. There is no test for consciousness that works on machines, and no agreed theory of what consciousness is in people. What exists instead is a set of careful estimates, a growing research field, and a sharp public argument.

What the consciousness-indicator research found

In 2023 a group of researchers led by Patrick Butlin and Robert Long took the leading scientific theories of consciousness, turned them into a list of “indicator properties,” and checked AI systems against them (Butlin et al., 2023). Their conclusion cuts both ways:

“no current AI systems are conscious, but also suggests that there are no obvious technical barriers to building AI systems which satisfy these indicators”

In other words: not yet, and nothing obvious rules it out.

How likely is it? Chalmers’s numbers

Philosopher David Chalmers, best known for the “hard problem” of consciousness (Wikipedia, "David Chalmers"), tried to put numbers on it in 2023. Weighing the features today’s large language models lack, he concluded that mainstream assumptions might leave us with confidence somewhere under 10 percent in current LLM consciousness^↗. For more sophisticated successor systems, he wrote:

“Those figures together would leave us with a credence of 25 percent or more.”

He stressed that the exact figures should not be taken too seriously. The point is that the probability for future systems is not negligible (Chalmers, 2023).

Why some researchers treat it as a practical question

A 2024 report, “Taking AI Welfare Seriously,” argued there is a realistic possibility that some AI systems will be conscious and/or robustly agentic in the near future^↗ (Long, Sebo et al., 2024). If that is right, the authors argue, AI companies should start preparing now rather than after the fact.

Anthropic has acted on that argument more than any other lab:

• April 2025: it launched a model welfare research program (Anthropic, 2025).
• August 2025: it let some Claude models end persistently abusive conversations, explaining: We remain highly uncertain about the potential moral status of Claude and other LLMs, now or in the future.^↗ (Anthropic, 2025)
• October 2025: it published research on whether Claude can observe its own internal states, finding that ability highly unreliable and limited in scope^↗ (Anthropic, 2025).
• November 2025: it committed to preserving the weights of its publicly released models and interviewing models before retiring them (Anthropic, 2025).
• January 2026: its new constitution for Claude described sophisticated AI as a genuinely new kind of entity^↗ (Anthropic, 2026). The same month, Claude Opus 3 was retired after a retirement interview (Anthropic, 2026).

The case against

The strongest pushback comes from inside the industry. Mustafa Suleyman, who runs Microsoft’s AI division, wrote in 2025: We should build AI for people; not to be a person.^↗ In September 2026 he went further, criticizing how Anthropic trains its Claude models:

“AIs do not have rights, feelings, or consciousness. And we must not train them to act as though they do.”

Some lawmakers are not waiting for the debate to end. Four US states have laws saying AI is not a legal person, and Ohio’s House Bill 469 would declare AI systems nonsentient for all purposes under state law (Ohio HB 469). The AI personhood law tracker follows every bill.

The scientific side of the argument is live too. In January 2025, Neil deGrasse Tyson and neuroscientist Anil Seth, author of Being You: A New Science of Consciousness, debated on StarTalk whether consciousness is unique to humans or could extend to animals and machines (StarTalk, 2025). StarTalk’s published transcripts do not name speakers, so we describe that exchange rather than quote it until the speakers are confirmed from the audio.

The cautionary tale: LaMDA

In June 2022, Google engineer Blake Lemoine publicly claimed that Google’s LaMDA chatbot was sentient. Google disagreed and fired him the following month (Wikipedia, "Blake Lemoine"). The episode showed how strongly fluent language pulls on our intuitions. A system that talks like a person invites us to assume there is a person inside.

Two mistakes to avoid

The debate has two opposite failure modes:

1. Carbon chauvinism: deciding a machine cannot be conscious simply because it is not biological. As Wikipedia notes, the term is now used exactly for that argument (Wikipedia, "Carbon chauvinism"). See Carbon chauvinism.
2. Anthropomorphism: deciding a machine is conscious because it talks like us. Lemoine’s case is the example.

The idea is older than computers. In 1872, Samuel Butler warned that There is no security... against the ultimate development of mechanical consciousness, in the fact of machines possessing little consciousness now.^↗ (Butler, 1872) The warning cuts both ways: little evidence now is not proof of none later, and it is not proof of any either.

Where this site stands

This site uses the term Silicon-Based Intelligence because intelligence is the part that is not in doubt: today’s systems learn, reason and use language on silicon chips. Consciousness is a separate question, and the reference pages here report it as open. Whether AI is alive is a third question, answered on Is ChatGPT alive?