Microsoft AI CEO says AI threats are real, and Anthropic is making it worse
Sep 17, 2026 · 52m
More Technology podcasts · More Business podcasts
Summary
Mustafa Suleyman, CEO of Microsoft AI, discusses the company’s new "Humanist AI Code of Conduct" and argues that AI safety requires both alignment and strict containment to prevent autonomous agent risks. He critiques Anthropic’s approach to model welfare, warning that treating AI as conscious could undermine control and alignment. Suleyman advocates for practical industry standards, such as banning neural communication between models, while addressing the tension between rapid innovation and the need for regulatory oversight.
Topics discussed
Sponsorships: Engine, CrowdStrike, and Mint Mobile
Introduction: Mustafa Suleyman and the Humanist AI Code
Alignment vs. Containment: The shift in AI safety focus
The Hugging Face incident and multi-agent coordination
Prohibiting neural communication and enforcing standards
Regulatory frameworks and industry self-regulation
Releasing the Humanist AI Code of Conduct for consultation
Model welfare and the audience of AI training documents
US government response and antitrust exemptions
Sponsorships: Engine and CrowdStrike
Product liability vs. proactive regulation
Moving the debate from Twitter to Congress
Anthropic's model welfare and the 'consciousness' debate
Enforcing safety standards and Microsoft's platform role
Sponsorships: Square, Pipedrive, and IBM Quantum
China, open-source models, and the 'winning' metaphor
Addressing skepticism about AI safety slowdowns
Defining humanist superintelligence and control
Technical solutions: Real-time monitoring and chip-level controls
Conclusion: Call to action and future markers
Sponsorships: McDonald's and Blinds.com
More from Decoder with Nilay Patel
Listen ad-free on Castria