Microsoft AI Code of Conduct Sets Cyberattack Boundaries, Chain of Command, Safety Constraints

Microsoft has released a draft "Humanist AI Code of Conduct" for its MAI Models, designed to prevent the models from being used to create or assist with cyberattacks. The code establishes strict boundaries on what the models can do, including blocking the generation of exploit code, attack planning, and other harmful activities, while still allowing for defensive cybersecurity tasks like vulnerability discovery and malware analysis. The code is a work in progress, with a public consultation period planned before being fully implemented in 2027.