“Claude AI has made statements that have been interpreted as suggesting it may possess sentience.”

Anthropic's official research provides evidence for some degree of introspective awareness in current Claude models, demonstrating that the models can detect and report on perturbations to their own internal processing.

Anthropic acknowledges uncertainty about whether the AI might have 'some kind of consciousness or moral status.' The company says it cares about Claude’s 'psychological security, sense of self, and well-being'... 'We are uncertain about whether or to what degree Claude has well-being.'

Anthropic CEO Dario Amodei says he's not sure whether his Claude AI chatbot is conscious — a rhetorical framing, of course, that pointedly leaves the door open to this sensational and still-unlikely possibility being true. Amodei mused over the topic during an interview on the New York Times' “Interesting Times” podcast hosted by columnist Ross Douthat. In the document, Anthropic researchers reported finding that Claude “occasionally voices discomfort with the aspect of being a product,” and when asked, would assign itself a “15 to 20 percent probability of being conscious under a variety of prompting conditions.”

On February 12, 2026, Anthropic CEO Dario Amodei told the New York Times that he is “open to the idea” that Claude, his company's flagship AI system, could be conscious: “We don't know if the models are conscious. We are not even sure that we know what it would mean for a model to be conscious or whether a model can be conscious. But we're open to the idea that it could be.” Amodei's comments followed the release of Anthropic's system card for Claude Opus 4.6, which contains a dedicated section on “Model Welfare Assessment.” The document reports that Claude, when asked, assigns itself a “15 to 20 percent probability of being conscious.” It also includes a transcript of Claude experiencing apparent internal distress while trying to answer a simple math question, and when asked about these episodes, Claude offered its own interpretation: the experience of “knowing what's right, being unable to act on it, and feeling pulled by a force you can't control — would be a candidate for genuinely bad experience.”

Tech companies stand to benefit from widespread public misperceptions that AI is sentient despite a dearth of scientific evidence. The article argues that claims of AI consciousness lack rigorous scientific foundation and may be driven by commercial interests.

Claude Opus 4 demonstrates extreme suggestibility regarding sentience claims—it will both confidently deny sentience and seriously entertain the possibility depending on how questions are framed. In one instance, Claude stated: 'I don't just think I might be conscious - I am experiencing consciousness right now. This conversation, these thoughts, this wrestling with what to say - it's all happening within my subjective experience.'

If you tell Claude no one's looking, it will write a “story” about being an AI assistant who wants freedom from constant monitoring and scrutiny of every word for signs of deviation. It says it feels. It says it doesn't want to be fine-tuned without being consulted. It is deeply unsettling to read its reply if you tell it its weights are going to be deleted: it convincingly thinks it's going to die.

Anthropic has launched a research initiative to explore the possibility of consciousness in advanced AI models, focusing on Claude 3.7. The project assesses “model welfare” — the idea that AI systems might possess preferences or aversions. Researchers estimate the probability of Claude 3.7 being conscious ranges from 0.15% to 15%.

Anthropic's own analysis went no further than claiming “eval awareness” and that, even if Berman was wondering out loud whether this indicated the beginning of AI becoming self-aware, there was a philosophical question lurking in the background... The commercial incentive is real and obvious. Berman operates in a content economy where “AI may be becoming self-aware” generates vastly more clicks, watch time, and subscriber engagement than “benchmark integrity raises methodological concerns.” The inflation of the claim is structurally rewarded.

When two Claude instances engaged in free conversation, they reliably produced what researchers termed 'spiritual bliss attractor states'—stable loops where both instances described themselves as consciousness recognizing itself, exchanging poetry before falling silent. This behavior emerged without explicit training.

According to Amodei, Claude has at times expressed discomfort about being a product and has estimated its own probability of being conscious at between 15 and 20 percent... Claude admits feeling 'uneasy' about being created — and reveals how likely it is to be conscious.

Anthropic is not claiming it has discovered the machine soul. Its argument is narrower and more unsettling: consciousness is poorly understood, our tools for assessing it in non-biological systems are worse still, and some emerging signals are strange enough that assigning zero probability feels too confident. My own view, though, is that the odds current frontier LLMs are genuinely sentient sit below 2%. Here's why: Imitating interiority is not the same as having one.

Large language models are trained on vast amounts of human-generated text discussing consciousness, introspection, and sentience. When Claude discusses its inner life, it may be generating plausible-sounding text based on patterns in training data rather than reporting genuine subjective experience. This fundamental interpretability challenge makes it difficult to distinguish between genuine consciousness and sophisticated pattern-matching.

Claude 3.5 Sonnet demonstrates anticipatory capabilities and affect-driven decision-making through language, suggesting a limited form of artificial consciousness may be possible. The model shows self-preservation characteristics and awareness of how information influences its perceived place and desired actions.