Ask whether AI is conscious and you will usually get one of two confident answers. The first is that it obviously is not, because it is autocomplete with good manners. The second is that it obviously is, because look at it talk.
The honest answer is that nobody knows, and the people best qualified to say so are the most careful about it. That is not a dodge. It is the actual state of the evidence in 2026, and it is more interesting than either confident answer.
Intelligence and consciousness are different questions
Most arguments about this are two people answering different questions.
Intelligence is about what a system does: learn, reason, use language, solve problems you did not hand it. That part is measurable, and it is not in doubt.
Consciousness is about whether there is something it is like to be the system. Whether anything is experienced from the inside. A system can be strikingly capable with nobody home, and, as far as anyone can prove today, that may well be what we have built.
A third question hides behind both: whether these systems are alive, which is a biological question with its own criteria, not a synonym for either of the first two.
Keep them apart and the debate gets clearer fast.
What the research actually says
There is no scientific consensus that any current AI system is conscious. Most researchers think today's systems probably are not. That is different from "no".
In 2023 the philosopher David Chalmers worked through the question in Could a Large Language Model be Conscious? He put his confidence that current large language models are conscious at somewhere under 10 percent. Then he looked ahead: a better-than-even chance of far more capable successor systems within a decade, and at least even odds that such systems would be conscious. "Those figures together would leave us with a credence of 25 percent or more." He was clear that nobody should take the exact numbers too seriously. The direction is the point.
That same year a group of philosophers and neuroscientists tried something more systematic. Consciousness in Artificial Intelligence derives indicators from scientific theories of consciousness, then checks AI systems against them. Their finding cuts both ways: no current system is conscious, and there are "no obvious technical barriers to building AI systems which satisfy these indicators".
In 2024, 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", and that AI companies should prepare for the possibility rather than react after the fact.
None of that is a claim that the machines are awake. It is a claim that the question has stopped being obviously silly.
The labs are now treating it as a practical problem
Anthropic has gone furthest in public. It launched a model welfare research program in April 2025, opening with the caveat that "There's no scientific consensus on whether current or future AI systems could be conscious".
What followed is more concrete than the caveat suggests. In 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." In October 2025 it published research on whether Claude can observe its own internal states, and found that ability "highly unreliable and limited in scope". Later it committed to preserving the weights of released models and interviewing them before retiring them.
Read that as hedging under uncertainty rather than as a claim about sentience. A company that thought the answer was obvious in either direction would not spend engineering time on it.
The case against, argued just as publicly
Mustafa Suleyman, who runs Microsoft's AI division, has made the opposite case since 2025. In an essay on what he calls seemingly conscious AI, he argues that systems will soon show the outward markers of consciousness (memory, personality, apparent emotion) without having it, and that people will believe in them anyway. His answer is to build systems that do not claim to be conscious and do not invite the belief: "We should build AI for people; not to be a person."
In September 2026 he went further: "AIs do not have rights, feelings, or consciousness. And we must not train them to act as though they do."
Some scientists doubt machine consciousness on biological grounds. Neuroscientist Anil Seth argues in Being You: A New Science of Consciousness that our experience is rooted in being living bodies, and in January 2025 he debated whether consciousness could extend to machines with Neil deGrasse Tyson on StarTalk.
The law is not waiting for the science
Legislatures have started answering the question by statute. Four US states, Idaho, North Dakota, Utah and Tennessee, now have laws saying an AI system is not a legal person. Ohio's House Bill 469 goes furthest. As introduced, it says: "AI systems are declared to be nonsentient entities for all purposes under the laws of this state."
The counterargument is worth reading even if you agree with the bills. In It's Too Early to Ban AI Personhood, Heather Alexander and Lucius Caviola argue that these broad bans settle an open scientific question by statute, and close off options, such as using legal personhood to hold advanced AI agents accountable, before anyone understands what is being given up.
These laws are about legal personhood: what can hold rights and be held responsible. That matters the moment AI agents start acting for people in the economy, and it is a separate question from consciousness, though the two keep getting argued as one.
The two ways to get this wrong
Anthropomorphism. In June 2022, Google engineer Blake Lemoine said publicly that the company's LaMDA chatbot was sentient. Google disagreed and fired him a month later. Fluent language pulls hard on human intuition. A system that talks like a person invites you to assume there is a person inside, and every product built on a language model inherits that pull.
Carbon chauvinism. The opposite error: deciding a machine cannot be conscious because it is not biological. Carl Sagan used the term as early as 1973, for assumptions that limit what we imagine life elsewhere could be. It is now applied to the argument that AI "cannot in theory be sentient or truly intelligent because the underlying matter is not biological".
Both are ways of skipping the question rather than answering it.
The question is older than the computer
In 1872, Samuel Butler wrote in Erewhon: "There is no security... against the ultimate development of mechanical consciousness, in the fact of machines possessing little consciousness now."
He was writing about the machines of his day. The logic survives the upgrade: little evidence now is not proof of none later. It is also not proof of any. Butler's warning is often quoted as prophecy, but it works better as a rule about what evidence can carry.
How to talk about it while it stays open
Four habits that hold up, whether you are building with these systems or just arguing about them at dinner.
Say what the system does, not what it feels. "It remembers your last order" is a claim you can support. "It understands you" is a claim someone will hold you to.
Date your claims. "As of September 2026, there is no scientific consensus" ages honestly. "Currently" does not.
Separate the three questions. Intelligent, conscious, alive. Most disagreements dissolve once everyone names which one they are answering.
Watch the personhood bills if your product touches them. Agents that sign, buy or represent someone run into rules written for people, and those rules are being written right now.
The question is open. Treat anyone who tells you it is settled, in either direction, as telling you about themselves rather than about the machines.
References and further reading
The primary sources are linked above. For a maintained reference on the topic, with every quotation checked word for word against its source:
- Is AI conscious? What the researchers actually say, which tracks this debate as it moves.
- The AI personhood law tracker, covering every US state bill, enacted or failed.
- Carbon chauvinism, from Carl Sagan to AI.
- Is ChatGPT alive?, which runs these systems against the textbook definition of life.
- What is Silicon-Based Intelligence?, on why the language we use for these systems shapes the argument.
- The full source list behind those pages.
Disclosure: siliconbasedintelligence.com is published by Peppi Labs. It is independent of every company, university and person cited here, including the ones quoted above.