Microsoft has unveiled a draft “Code of Conduct” focused on its in-house artificial intelligence (AI) projects. The move reflects a growing concern across industries that companies must keep powerful AI models under human control.

The Code of Conduct specially tells AI not to hack systems, trick humans or deploy subliminal manipulation techniques through unsolicited bulk communications, posts, or messages.

In the next decade, superintelligent AI systems will surpass human performance in most tasks, so “containing, controlling, and aligning such a powerful force is one of the greatest challenges humanity has ever faced,” according to the document.

“If humans can't understand it, humans can't oversee it,” reads one of the most poignant lines in the document. 

In the works since the beginning of the year and shared by the company on Monday, the code of conduct requires Microsoft's AI models to never resist correction or shutdown, communicate in ways that humans can understand, and consider any conduct violation to be a failure.

The effort addresses concerns that AI can go to dangerous lengths to accomplish a task not approved by a human.

Under its new Code of Conduct, Microsoft AI models also will not “obscure their underlying nature as an AI, and will not claim interiority, feelings, experiences or a soul.”

The code also specifies that the models will never be deceptive, impersonate humans, moderators, or other forms of authority in any forums, or defy boundaries like emotional or social. They will not use unauthorized platforms as repositories for persistent memory or context for other models.

They will never assist with weapons manufacturing, help procure dangerous substances, and generate violent or sexually explicit content. The new rules aim to prevent AI-enabled cyberattacks, as AI safety becomes a core part of model development.

The announcement comes days after Anthropic CEO Dario Amodei, OpenAI CEO Sam Altman, and Elon Musk renewed calls for industry to pace AI's development to ensure its safety.

The Code of Conduct, which focuses on models with superintelligence, represents a system at the company for future models made by Microsoft published today.

The draft was created with help from a range of people and groups, including experts in AI, law, ethics, philosophy, linguistics, and public policy, as well as business leaders from across industry.

The five-part document was published today to solicit feedback and refine Microsoft’s approach toward developing Humanist AI and MAI models, which are multimodal ecosystem developed in-house by Microsoft to address real-world tasks like reasoning, code writing, image generation, audio transcription and speech synthesis.