AI Summary
5 min readMicrosoft has released a new code of conduct for its AI models, and the host of Hard Fork AI welcomes the transparency. The code echoes the Anthropic Constitution in its aim to make explicit what the models are supposed to do and not do. The host argues that this kind of public commitment is valuable because it gives observers something to hold companies accountable to, especially as other labs may have quietly deprioritized safety work amid the race to ship.
The central principle: AI must be subordinate
Mustafa Suleyman, head of Microsoft AI, summarized the entire document in five words: "people matter more than AI." The host endorses this framing, comparing it to how aviation safety doesn't stop us from flying but makes flying safe. The post argues that fears about AI—swarms of agents breaking out of sandboxes, unauthorized hacks, agents modifying their own logs—are real but should be met with guardrails, not a halt to development. The host contrasts this with what he sees as hyperbolic doomsaying from some labs, noting that Microsoft's approach is to say "we see the risks, and here is how we will manage them" rather than "we must slow everything down."
The ten-point conduct code
The host walks through several of the ten principles Suleyman laid out:
Continue reading the full summary in the app — free to try.
Read Full Summary →Free • No credit card required
Never miss an episode of Hard Fork AI
Get every new episode summarized in your inbox — free, ~5 minutes to read.
No spam. Unsubscribe anytime.
What you'll learn
- 1 (01:10) **Microsoft's AI Code of Conduct Announced** - The host introduces Microsoft's new AI code of conduct, comparing it to Anthropic's constitution and praising the transparency of publishing explicit AI governance rules.
- 2 (01:56) **Mustafa Suleiman's Post and the "Approaching the Frontier" Framing** - Suleiman's post outlines the urgency behind the code and draws criticism for positioning Microsoft as "approaching the frontier" while others are already there.
- 3 (04:52) **Host's Positive Framing of Guardrails vs. Doom Scenarios** - The host argues that AI labs should focus on safety guardrails rather than apocalyptic warnings, using the analogy of airplanes.
- 4 (05:25) **Principle 1: People Matter More Than AI** - The first and most foundational principle of Microsoft's code is that human welfare takes precedence over AI capabilities.
- 5 (05:45) **Principle 2: No Model Welfare or AI Rights** - Microsoft explicitly states that AI should not have legal personhood or rights, a stance the host strongly endorses.
- 6 (06:15) **Principle 3: Never Violate the Code, Even to Finish a Job** - The code is hard-coded into the model's system prompt, requiring it to refuse tasks that would break the rules.
- 7 (06:45) **Principle 4: Interruptible, Correctable, Shut Down-able** - Microsoft pledges not to ship any model that cannot be reliably shut down or corrected by humans.
+ Full timestamped outline available in the app
Guests on this episode
Show Notes
In this episode, we explore Microsoft's recently released code of conduct for their AI models, highlighting its emphasis on transparency and accountability. The episode discusses the principles outlined in this code and their implications for AI development and ethical considerations.
Chapters
00:00 Introduction to Microsoft's Code of Conduct
02:01 Purpose of AI in Service of People
04:01 Key Principles of the Code
05:58 Transparency and Accountability
08:01 Implications for AI Development
10:02 Conclusion and Future Outlook
Show Links
More from this podcast
Hard Fork AI →