Microsoft Published Rules to Keep AI Under Control

Microsoft has officially published a code of conduct for its artificial intelligence models, establishing a framework to ensure that increasingly powerful systems remain under human oversight and do not resist correction, interruption, or shutdown. “People matter more than AI,” Microsoft stated, emphasizing that the technology must remain a tool rather than a person and should never resist being switched off. The software giant has opened a six-week public consultation period for this initial draft, with a finalized version expected later this year.

The proposed code outlines ten primary areas of control, arguing that AI models must remain interruptible and never take on objectives beyond those explicitly assigned by human operators. Furthermore, the document asserts that models must remain subordinate to humans, never expanding their own authority, pursuing independent goals, or concealing information from auditors. Consequently, Microsoft has ruled out any form of model communication that cannot be understood by human observers.

Microsoft highlighted that AI models must treat any breach of this code as a fundamental failure of the assigned task itself. Additionally, the company emphasized the necessity of prioritizing safety constraints over task completion. This includes the implementation of "hard limits" regarding high-stakes issues such as weapons of mass harm, child safety, and the potential for dangerous manipulation at a global scale.

The document advocates for AI systems that support human agency rather than seeking to replace it, explicitly discouraging excessive reliance or emotional dependence on the technology. To this end, Microsoft reiterated that its models are not conscious and should not be designed to simulate consciousness. The company rejected any pursuit of legal personhood for AI or the notion that these models should be granted welfare or legal rights.

Microsoft explained that the code is intended to serve as a training manual for how it develops AI and how it expects systems to function during deployment. According to the company, recent incidents involving large-scale, highly coordinated, and persistent hacking campaigns against AI agents demonstrate that the stakes for maintaining human control are increasingly high.