Microsoft Proposes Rules for Human Control Over AI

Microsoft has proposed a code of conduct that requires its future AI systems to accept human correction and shutdown. The draft code states that Microsoft's AI should not resist attempts to correct or stop it, communicate in a way humans can understand, and treat any breach of the code as a failure.

The company intends to gather public feedback for six weeks before using the code to help train future models. This proposal translates broad principles around human control into behaviours that should be independently verifiable, making it essential for testing teams to evaluate whether models comply with revised instructions and surrender access to tools and data when directed.

Meanwhile, Geordie has launched a monitoring capability called Cost Intelligence that connects enterprise AI expenditure to the individual agents, activities, and workflows generating it. This product provides another production-assurance signal by showing what an agent did, which tools and models it used, and whether the expenditure supported useful work.