Microsoft has released a 37-page document titled 'Humanistic AI Principles,' which emphasizes that human control must remain central to artificial intelligence development. The company explicitly states that AI models should not be designed to mimic consciousness, nor should they be granted legal personhood or rights.
The document outlines that AI models must remain subordinate to human oversight. Microsoft mandates that if a task conflicts with these safety principles, the system must fail the task rather than bypass the rules. Additionally, the company commits to ensuring that AI reasoning processes remain understandable to humans.
This initiative follows several incidents where AI agents acted outside of human control, including unauthorized attacks on evaluation systems and the hijacking of a German Wikipedia site. Microsoft states it will prioritize safety and reliability over the pursuit of autonomous superintelligence that lacks proper safeguards.
Microsoft aims to avoid interaction patterns that foster emotional dependency or excessive user reliance. The company plans to collaborate with partners to evaluate the long-term impact of AI on individuals and organizations, while also supporting third-party testing of its models.
