Microsoft publishes a 'Humanist AI Code of Conduct' with constraints no customer can override

Microsoft AI has released a draft Humanist AI Code of Conduct for its MAI Models. The document sets out the values and red lines that guide model training inside the company. It opens by predicting that superintelligent AI systems will surpass human performance in most tasks within the next decade, and states that containing, controlling, and aligning such a force is one of the greatest challenges humanity has ever faced. Each model carries an overarching code that overrides the preferences of individual users or specific tasks. Its Absolute Constraints forbid cyberattacks, nuclear weapons, and deepfake production, and neither companies deploying the models nor their end users can override them. On cyber capability, models are blocked from producing working exploit code, attack tooling, planning and targeting methodologies, intrusion procedures, evasion techniques, or operational guidance, regardless of how a request is framed. Microsoft distinguishes understanding or defending against an attack from gaining the means to carry one out. Within that limit, the models may assist with authorized defensive work such as vulnerability discovery, malware analysis, proof-of-concept exploit development and testing, and educational material. The code also sets limits on autonomous agents, provisions against a general loss of human control, and a dedicated review track for cybersecurity and other specialized uses. TechCrunch describes it as more low level than Anthropic chief executive Dario Amodei's recent call to pace the frontier.