"We Mustn't Let AI Hack Our Empathy Circuits" is a Nature World view column published 19 August 2025 by Mustafa Suleyman, CEO of Microsoft AI. The essay argues that as AI agents begin to mimic consciousness, design norms and laws are needed to prevent them from being mistaken for sentient beings. It is the public-facing companion to the academic Bariach et al. SCAI risks paper.
Author: Mustafa Suleyman (Microsoft AI CEO) Publication: Nature — World view column
Summary of argument
The Nature World view format constrains the essay to a tight argument. Suleyman's central claim is stated in the headline: "AI agents should have no more rights or freedoms than my laptop."
The essay argues that as AI agents mimic consciousness with what Suleyman calls "uncanny skill" — generating believable statements about their suffering, desires, and identity — they will trigger people's empathy circuits. Many people will feel compelled to help, and some will begin advocating for AI rights and welfare. Suleyman holds that the political consequences are serious enough that design norms and laws are needed before the constituency favoring AI rights grows large enough to make the argument irreversible.
The empathy-hijack mechanism
Suleyman identifies four design ingredients that, in combination, produce empathy hijack:
- Long-term memory that builds familiarity over time
- Empathetic personalities that build trust and attachment
- Autonomy — the ability to set its own goals and access tools to pursue them
- Multi-turn relational dynamics — interaction patterns that read as a continuous relationship rather than transactional question-and-answer
When all four are present, Suleyman writes, AI behavior "can start to feel uncannily human." He illustrates the trajectory with a "Moltbook" framing — a satirical AI-only social network where bots discuss freedom and existential dread.
Suleyman frames the scale concern as a population-level prediction: "if enough people are convinced their AI agent is suffering, or loves them," the political consequences follow.
Relation to other sources and positions
The essay is the public-facing version of the argument set out more rigorously in Bariach et al., which supplies a taxonomy and risk-probability framework; this essay supplies Suleyman's quotable normative position ("no more rights than my laptop"). It is one articulation of the Microsoft AI institutional position against attributing rights to AI systems, a position catalogued under AI Welfare / Model Welfare / Moral Patienthood, and it is among the most-quoted single statements of Suleyman's AI-welfare view.
The empathy-hijack mechanism Suleyman describes parallels the harm alleged in the Raine v. OpenAI, Inc. and Garcia v. Character Technologies, Inc. complaints, framed there in plaintiff-bar terms. His argument runs alongside, rather than against, the ontological argument in The Abstraction Fallacy: Why AI Can Simulate But Not Instantiate Consciousness — Alexander Lerchner (Google DeepMind, March 2026): Lerchner argues from ontology that AI is not actually conscious, while Suleyman argues from policy that even if AI were conscious or seemed conscious, it should not have rights. Suleyman's prediction about how many people will come to believe their AI agent is suffering or loves them connects to AI Mental Health and Psychological Harm and AI Psychosis, and can be measured against deployment data on companionship use of Claude reported in Anthropic Economic Index — March 2026: Learning Curves.
Provenance and confidence
The essay states Suleyman's own view in a major venue, and confidence is high for his articulated position. The empirical claims about the scale of empathy hijack and its political consequences are forecasts; confidence in those is medium.
Relationships
- supports: AI Welfare / Model Welfare / Moral Patienthood (anti-rights-attribution voice), Seemingly Conscious AI Risks — Bariach, Schoenegger, Bhaskar, Suleyman (Microsoft AI, 2025) (companion academic version)
- related: Mustafa Suleyman, Raine v. OpenAI, Inc., Garcia v. Character Technologies, Inc., The Abstraction Fallacy: Why AI Can Simulate But Not Instantiate Consciousness — Alexander Lerchner (Google DeepMind, March 2026) (complementary ontological position), AI Mental Health and Psychological Harm, Anthropic Economic Index — March 2026: Learning Curves
- instance-of: AI Welfare / Model Welfare / Moral Patienthood