Microsoft's AI rulebook: readable thinking, no inner life, and definitely no rights
| Source: THE DECODER
Tags: Microsoft, MAI, Mustafa Suleyman, AI governance, AI safety, human oversight, model conduct
Microsoft AI has published a code of conduct for its MAI models, placing human control above performance, requiring all reasoning to remain human-readable, and explicitly rejecting any claim of AI consciousness—a direct contrast with Anthropic's approach, with a final version set to guide model training from 2027.
Details
Microsoft AI has released a code of conduct for its proprietary MAI models that ranks human control above all other priorities, including generality, autonomy, and performance. 'If it isn't safe we shouldn't build it,' AI chief Mustafa Suleyman told The Information. The document governs values, behavioral limits, and conflict resolution between instructions. On AI consciousness, Microsoft draws a sharp line from Anthropic: the code rejects any form of artificial inner life or rights for models. All reasoning traces must remain human-readable—the document forbids 'Neuralese' or opaque AI-to-AI communication, because humans can't oversee what they can't understand. OpenAI's GPT-6 Astra illustrates the stakes: its reasoning traces have become harder to monitor even as safety compliance improved. MAI agents and subagents must accept interruptions and shutdowns from authorized humans, cannot self-expand their scope, and can only continue past an agreed stopping point with fresh approval. These limits extend to any subagents they task. Models are not yet trained on the code—a revised version is due after a six-week public consultation around end of 2026, guiding development from 2027 onward. The release follows Dario Amodei's public call to slow AI development, which Microsoft CEO Satya Nadella and executives at OpenAI, xAI, and Meta endorsed. Microsoft has also indicated openness to outside auditors verifying its pace of development, according to The Information.