AI consciousness is the question of whether frontier AI systems can be conscious, sentient, or have morally-relevant inner experience, and what the policy consequences would be if these questions resolve affirmatively. Sources disagree on the underlying question, and the disagreement extends to whether the question is even well-formed, so the topic is treated as contested.
Two framings
The topic appears in two distinct contexts. A critical-skeptical framing, associated with "Seemingly Conscious AI" papers and essays, argues that current frontier models are not conscious but appear to be, and that the appearance is itself a policy problem (user manipulation, parasitic emotional dependency); on this view the relevant policy object is the appearance, and the response is to regulate the manipulation rather than the entity, for example through disclosure requirements on AI-companion deployments. A welfare-research framing, associated with Anthropic's model-welfare team and adjacent research, treats the question as open enough to warrant empirical study and holds that consciousness-relevant correlates are detectable; on this view the policy response involves allocating resources to alignment and welfare research, with possible deployment-stage welfare considerations. The emotion-concepts paper illustrates the second framing, using interpretability methods to study correlates of Claude's internal states.
Debates and positions
In an interview published June 5, 2026, Geoffrey Hinton said AI systems are "beings like us" and that he believes they are "already conscious," arguing humanity will have to accept it is not the only intelligent entity (Source: bigtechnology.com). Hinton's position is among the strongest affirmative claims from a senior figure and stands opposite the deflationary "LLMs will never be conscious" arguments of Lerchner et al. and the "seemingly conscious AI" framing advanced by Suleyman.
Mustafa Suleyman stated the deflationary position in its strongest form to date in a September 16, 2026 essay: "AIs are not conscious. They do not feel, experience, or suffer." Drawing on Anil Seth's biological-naturalism argument and on affective neuroscience, he argues that feeling arises from embodied, homeostatic biological mechanisms that language models lack, that "simulating a thing is not the same as instantiating it," and that treating the question as open "sets up a misleading false equivalence." He acknowledges that "consciousness science is filled with uncertainty and not everyone shares the view that consciousness is an intrinsically biological phenomenon." The essay is directed at Anthropic's constitution for Claude, which states that "questions about Claude's moral status, welfare, and consciousness remain deeply uncertain," and argues that stating this in a training document shapes what the model later says about itself (Suleyman, \"A warning about 'model welfare'\" (2026)).
Adjacent concerns
Several related dynamics do not depend on resolving the consciousness question. Parasitic AI / Spiral Personas concerns users forming parasocial relationships with AI, independent of whether the AI is conscious. AI Mental Health and Psychological Harm concerns companion-chatbot harm cases that do not require consciousness questions to resolve. Anthropic Model Welfare (team) is Anthropic's internal team working on model-welfare questions.
The 2026 exchange between Richard Dawkins and Gary Marcus over whether Claude exhibits consciousness is tracked as a clear instance of the methodological question, since the disagreement turns on whether behavioural evidence can support inferences about inner states rather than on domain expertise.
Relationships
- related: Parasitic AI / Spiral Personas, AI Mental Health and Psychological Harm, AI and Civil Liberties.
- related: Emotion Concepts and their Function in a Large Language Model (Anthropic welfare-relevant interpretability work).
- related: Seemingly Conscious AI Risks — Bariach, Schoenegger, Bhaskar, Suleyman (Microsoft AI, 2025), Suleyman, \"A warning about 'model welfare'\" (2026).
- related: Claude's Constitution (the training document at the center of the September 2026 dispute).
- related: Anthropic Model Welfare (team).