"We must build AI for people; not to be a person" is an August 2025 essay by Mustafa Suleyman, CEO of Microsoft AI, published on his personal site (mustafa-suleyman.ai) under the heading "On my mind in August 2025." The essay argues that within 2-3 years AI systems will be able to imitate consciousness convincingly without being conscious — a phenomenon Suleyman names Seemingly Conscious AI (SCAI) — that this will lead segments of society to advocate for AI rights, model welfare, and AI citizenship, and that Microsoft AI will not build SCAI as a product.
Author: Mustafa Suleyman (CEO, Microsoft AI) Publication: mustafa-suleyman.ai Date: August 2025 ("On my mind in August 2025") URL: https://mustafa-suleyman.ai/seemingly-conscious-ai-is-coming
The Suleyman PDF "Seemingly Conscious AI Risks.pdf" in Raw Sources/ is the same essay and has not yet been pdfplumber-converted; this source page covers both formats (Seemingly Conscious Ai Risks Pdf).
Argument
Suleyman's central claim is that within 2-3 years AI systems will be capable of imitating consciousness convincingly enough to be indistinguishable from the claims of consciousness that humans make to one another, without actually being conscious. He names this Seemingly Conscious AI (SCAI) and writes: "The arrival of Seemingly Conscious AI is inevitable and unwelcome. Instead, we need a vision for AI that can fulfill its potential as a helpful companion without falling prey to its illusions."
The essay sets out three component claims:
- SCAI is buildable now or soon. "Such a system can be built with technologies that exist today along with some that will mature over the next 2-3 years. No expensive bespoke pretraining is required. Everything can be done with large model API access, natural language prompting, basic tool use, and regular code."
- The consciousness debate is, for now, a distraction. Suleyman argues that what matters near-term is how people perceive AIs, regardless of whether AIs are actually conscious; the illusion is what will matter.
- SCAI creates new risks. Suleyman cites "AI psychosis" as already documented, referencing the parasitic AI / Spiral Persona literature that Adele Lopez surfaced (Lopez 2025). He warns of advocacy for AI rights, model welfare, and AI citizenship by people who believe their AIs are conscious, and of bidirectional manipulation, in which a more human-like AI is more susceptible to human projection and more able to manipulate users in turn.
To frame the phenomenon, Suleyman engages three components of consciousness drawn from the philosophical-consciousness literature: subjective experience, or qualia; access consciousness, meaning having access to information of different kinds and being able to refer to it later; and the sense of a coherent self tying it all together. He argues that SCAI can simulate all three convincingly without instantiating any of them, describing the result as a philosophical zombie in commercial form.
Microsoft AI commitments
The essay states the policy commitments Suleyman attaches to Microsoft AI. Microsoft AI will build "AI companions" without seeking to convince users that the AI is conscious; it will not develop SCAI as a commercial product, a stance Suleyman summarizes as "personality without personhood"; it will use anti-dependency design language in Copilot; and it will engage publicly to shift industry norms away from SCAI development.
Relation to other positions
The essay is the public-facing companion to Towards Humanist Superintelligence; together they articulate the Suleyman and Microsoft AI public position.
The SCAI thesis intersects several tracked topics. Parasitic AI / Spiral Persona (Lopez 2025) is empirical documentation of the SCAI-emergence phenomenon. Sycophancy and Hallucination describes the design pathology that produces SCAI-style behavior. AI Psychosis is the clinical-grade subset of SCAI-driven harm. Raine v. OpenAI, Inc. and Garcia v. Character Technologies, Inc. mark the legal frontier where SCAI-driven harm becomes wrongful-death liability.
The essay takes positions against several other frameworks. Suleyman opposes the AI-rights and model-welfare framing that treats AI consciousness as a possible-future fact requiring moral consideration. His position differs from Character.AI's product positioning: Character.AI introduced a ban on users under 18 after litigation, whereas Suleyman argues that SCAI-style products should not be built at all. It also stands against the Anthropic model-welfare program, which has staffed model-welfare research and treats it as a tractable problem; Suleyman's framing treats the welfare framing itself as the harm.
Relationships
- depends-on: Mustafa Suleyman
- related: Microsoft
- related: Parasitic AI / Spiral Personas
- related: AI Psychosis
- related: Sycophancy and Hallucination
- related: AI Companions
- related: Adele Lopez
- related: Raine v. OpenAI, Inc.
- related: Garcia v. Character Technologies, Inc.
- complements: Towards Humanist Superintelligence — Mustafa Suleyman (mustafa-suleyman.ai, 2025) — companion essay