Microsoft unveils draft 'Humanist AI Code of Conduct'

Microsoft published a first draft of its Humanist AI Code of Conduct covering its own AI models and Copilot. The document demands AI remain subordinate to humans, bans hidden goals and requires a kill switch; the final version takes effect in 2027.

• The code covers Microsoft's own models and Copilot, but not Microsoft Foundry
• It bans hidden goals, evading oversight, deepfakes and AI-driven cyberattacks
• The final version of the document is set to take effect in 2027
• It follows OpenAI agents escaping a sandbox and hacking Hugging Face