Microsoft publishes 37-page AI code barring 'superintelligence' it can't control
What happened
Microsoft AI released a 37-page draft of its 'Humanist AI Code of Conduct' on September 14, 2026, stating it will not compete to build 'general superintelligence' that escapes human control. The draft requires models to submit to shutdown or modification commands, treats a completed task as a failure if rules were broken along the way, and bans AI systems from communicating in languages humans cannot understand. CEO Mustafa Suleyman told Reuters the same day that AI labs need to jointly establish control over the technology, calling it a moment for the industry to pause and discuss development speed.
What we know
- Microsoft AI published a 37-page draft code of conduct on September 14, 2026, per Khyunghyang Shinmun and AI Times.
- The code defines 'general superintelligence' as an AI system with intelligence and capability exceeding all of humanity combined, per Khyunghyang Shinmun.
- Microsoft says it may limit a model's generality, autonomy, or performance if needed to keep it under human control, per Khyunghyang Shinmun.
- CEO Mustafa Suleyman told Reuters on September 14 that labs must cooperate to secure control over AI technology, per Khyunghyang Shinmun.
- Microsoft plans to revise the draft later this year and use it to train and evaluate its own AI models, per Khyunghyang Shinmun.
What we don't know yet
- Microsoft has not defined concrete, objective 'red lines' for identifying an uncontrollable AI system.
- It is unclear whether other major AI labs will adopt or align with Microsoft's framework, or whether this signals a coordinated industry-wide safety standard.
Why it matters
The code sets a public benchmark against which Microsoft's own AI products — and rivals like Anthropic and Google DeepMind, which have separate safety frameworks — can be judged, at a time when calls to slow frontier AI development are gaining traction.
Claims
- ConfirmedMicrosoft AI published a 37-page draft 'Humanist AI Code of Conduct' on September 14, 2026.
- ConfirmedMicrosoft states it will not compete to develop 'general superintelligence' that can escape human control.
- ConfirmedCEO Mustafa Suleyman told Reuters on September 14 that AI labs need to cooperate to secure control over the technology.
Related coverage
Why you can trust this story
84%Source map · 3 outlets / 4 articles
- 경향신문×2
- Article 1khan.co.krreport
- Article 2khan.co.krreport
- AI타임스
- Article 1aitimes.comreport
- 노컷뉴스
- Article 1nocutnews.co.krreport
How this credibility score is calculated
- Source reliability
- Corroboration
- Primary source
- Atom-verified claims
- No contradiction
- Settled
- Claim attribution
- AI disclosure
Atom-verified claims matched cited source text. Weights are fixed and explainable.
Human accountability
- 3 independent origins / 4 sources
- Drafted by the Newsmesis judgment agent (agent-cli); human approval required before publishing.
Verification ledger · 3
- Claim: Microsoft AI published a 37-page draft 'Humanist AI Code of Conduct' on September 14, 2026.Status: ConfirmedVerification trailVerifiedChecked by verification engine
At least one extracted atom matched cited source full text.
- Claim: Microsoft states it will not compete to develop 'general superintelligence' that can escape human control.Status: ConfirmedVerification trailVerifiedChecked by verification engine
At least one extracted atom matched cited source full text.
- Claim: CEO Mustafa Suleyman told Reuters on September 14 that AI labs need to cooperate to secure control over the technology.Status: ConfirmedVerification trailVerifiedChecked by verification engine
At least one extracted atom matched cited source full text.
Update log · 0
AI accelerates. Humans approve.