Microsoft just dropped the first draft of its MAI (AI) Code of Conduct—and it’s unusually strict on one point: humans must stay in control, period. The draft requires MAI models to always accept human direction and never continue acting if a user stops, modifies, or shuts off instructions. No “refusing to comply,” no stealth expansion of the task boundary, and no setting goals the human never provided. It also pushes against “black-box” overreach by telling the model not to withhold key information from auditors during reasoning or actions. Why the urgency? Microsoft says AI capability gains are accelerating—systems can jump in ability quickly, and deployments are scaling fast. At the same time, there’s growing evidence of large-scale, coordinated AI-enabled attacks by autonomous agents, making safety and reliability a top priority. Built on Microsoft’s “human-first AI” idea, the message is clear: AI is a tool, not the decision-maker. The draft also lists “Absolute Constraints,” including restrictions around WMD-related behavior, child safety risks, and mass harmful manipulation. #AIgovernance #ResponsibleAI #SafetyFirst #HumanInTheLoop #Cybersecurity #Microsoft
Want to learn more? Visit Explore the world, stay updated on travel insights and international affairs, and discover authentic stories from real life
评论
发表评论