Microsoft proposes shutdown and correction rules for its future models

On September 14, Reuters reported that Microsoft released a draft code of conduct for 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 refusal to comply would be treated as a failure rather than a permitted autonomous objective. For legal-tech operators building on Microsoft infrastructure or selling into regulated legal environments, the draft supplies a potential procurement vocabulary for intervention, monitoring and fail-safe controls. What matters next is whether these principles become testable technical requirements and contract commitments rather than high-level assurances.