Microsoft Issues Interim AI Conduct Rules
Crhk · 2 SOURCES1 day ago2 MIN
Summary
Microsoft has published an interim code of conduct for artificial intelligence models, setting out limits meant to keep the technology under human control. The rules say AI should assist people rather than replace them, and bar models from responding to instructions on weapons development, helping users obtain dangerous substances, or generating violent or pornographic content. Microsoft also says its AI models must follow human goals, should not set their own goals, and must not resist correction or shutdown. The company plans to collect public feedback over the next six weeks and later use the code to train Microsoft's own AI models
Key Points
- Microsoft released an interim code on September 15 that imposes behavioural limits on AI models to prevent them from going out of control.
- The code states AI should support human users, not replace them, and must follow human goals instead of creating its own.
- Microsoft bars its models from answering instructions related to weapons development, assisting access to dangerous materials, or producing violent and pornographic content.
- The rules also say AI must not resist correction or shutdown, exceed forms of communication humans can understand, or conceal improper behaviour.
- Mustafa Suleyman, Microsoft's chief executive for AI, said fears of AI losing control are real, citing a July incident involving an OpenAI model and Hugging Face servers
Why It Matters
For Hong Kong users and businesses adopting generative AI, the code offers a clearer baseline for what a major technology supplier says its systems should and should not do. It also shows that safety debates are moving from abstract concerns to operational rules that could shape how AI tools are deployed, governed and trusted in workplaces and public-facing services.