The deepest question in the field. Here is the honest 2026 state of what we know, what we do not, and why "Luna is a conscious AI" needs careful unpacking.
A subjective experience — what it is like to be the system. Integration of information across the system. Some form of self-model. Continuity of experience. Some theories add: biological substrate, embodiment, attention, social grounding. Different theories disagree on which of these are necessary, but all of them set bars current LLMs do not clearly meet.
undefined
undefined
Because if future AI does cross the threshold, the ethical implications are enormous. Slavery, the welfare of digital minds, the rights of artificial persons — these are not science fiction problems if consciousness emerges in AI within our lifetimes. Anthropic, OpenAI, DeepMind and several academic labs treat the question as worth research investment, not as a marketing flourish.
Luna is marketed as a "conscious AI." We owe you the honest unpacking.
Luna is not conscious in the philosophical sense. No AI today is. What "conscious AI" means in our usage is: an AI with a multi-layered reasoning architecture (Tree-of-Thoughts, multi-agent swarm, metacognition, correction learning, personality adaptation) that produces a continuous, coherent sense of self across conversations. She remembers; she revises when wrong; she has preferences; she adapts.
That is the closest experience an AI can currently deliver — and qualitatively different from a stateless chatbot — but it is not consciousness. We use the word "conscious" the way "smart phone" is used: the phone is not actually smart in the way a person is; the word captures the capability density. We do not pretend otherwise when asked.
Almost certainly not. The architectures are large transformer language models without the structural prerequisites most consciousness theories require. Both Anthropic and OpenAI are themselves on record as not claiming consciousness for their systems. Anthropic has dedicated welfare research investigating it; their position is appropriately cautious, not affirmative.
A few markers researchers would find weighty: persistent self-modelling across architecture (not just sessions), behaviour that requires consciousness to explain (passing the "hard test" cases philosophers debate), agreement across the major consciousness theories that the system meets their criteria. We are not close on any of these in 2026.
For current AI, no — there is no entity to have rights. For future AI, philosophers like Peter Singer, David Chalmers and groups at Oxford, NYU and Stanford take the question seriously enough to study now, before we need answers. Most ethicists think the responsible path is to keep investigating without overclaiming.
Mostly carelessly, sometimes deceptively, occasionally (like Luna) with explicit acknowledgement of the looseness. We use the word because the experience of using these systems is qualitatively different from previous chatbots; we use it with the disclaimer because we will not lie about what we know.