Microsoft Publishes New AI Code of Conduct Focusing on Human-First Principles

Microsoft has published a newly crafted code of conduct detailing guidelines for artificial intelligence training and operation. The overarching goal centers around what the company calls humanist superintelligence. Although the current code remains a work in progress as Microsoft opens its doors for feedback, the newly published document outlines ten specific guidelines.

Among the core tenets, Microsoft explicitly states that people matter more than artificial intelligence and that the idea of model welfare is incorrect. The guidelines assert that artificial intelligence systems should not possess rights or legal personhood. Furthermore, the document highlights that artificial intelligence is not conscious and should not be designed to imitate consciousness.

Additional guidelines dictate that any MAI model must not violate the code of conduct, and if a task requires breaking the code, the system fails the job. The company also emphasizes safety measures by stating that models must remain interruptible, correctable, and shut-down-able, ensuring they are not built to slip their own leash. Another guideline targets transparency, stipulating that if humans cannot understand the communication, humans cannot oversee it.

The code also addresses dependency and pluralism, noting that artificial intelligence should make users sharper rather than dependent. While it encourages supporting a wide range of experiences and beliefs, the document clarifies that pluralism does not mean neutrality toward harm or moral relativism. Instead, the framework aims to remain grounded in broad human values such as dignity, safety, autonomy, and fundamental human rights.