Microsoft proposes correction and shutdown rules for future models

On September 14, Reuters reported that Microsoft released a draft code of conduct for its advanced AI systems with a six-week public-comment period, including expectations that systems accept correction and shutdown and communicate in understandable terms. Microsoft said a refusal to comply would be treated as a system failure, and the code is intended to inform training of future models. For KM teams operating heavily in Microsoft 365, the draft supplies useful governance language for intervention, monitoring and fail-safe behavior across agents that can reach organizational knowledge. The practical test will be whether these principles become auditable product controls and contractual commitments that customers can verify.