Microsoft's MAI code of conduct bans AI consciousness claims, demands human-readable reasoning, and puts human control above performance—diverging sharply from Anthropic's approach.
September 14, 2026
brightray analysis
Summary
Microsoft published a code of conduct for its MAI models that prioritizes human control above autonomy or performance, bans AI from using unreadable reasoning formats ("Neuralese"), and explicitly rejects any claims to AI consciousness, feelings, or rights. The document will guide model development starting in 2027 after a six-month public consultation. This contrasts with Anthropic's constitution for Claude, which treats AI subjective experience and moral status as open questions and researches "functional emotions."
Why it matters
- Microsoft's MAI code explicitly bans AI from claiming consciousness, feelings, or rights — a direct repudiation of Anthropic's open-question stance on AI moral status and functional emotions.
- Models must not use unreadable "Neuralese" reasoning in internal chains or agent-to-agent communication, because oversight requires human-legible traces.
- Human control is the top-ranked constraint: Microsoft says it will sacrifice generality, autonomy, or performance to maintain it — and is open to outside auditors verifying any self-imposed slowdown.
- The code will inform model development starting 2027, but only after a six-week public consultation and revision cycle — it is not yet used in training.
- Microsoft acknowledges readable reasoning traces are insufficient: a model's stated reasons need not reliably reflect its actual behavior, leaving the core control problem unsolved.
Community notes—
No notes yet — be the first.
See every signal in the Feed