Microsoft's MAI code: human control first, no inner life

Microsoft's code arrives amid a broader shift in how AI companies talk about speed. The release follows Anthropic CEO Dario Amodei's call to slow the industry's pace, which Microsoft CEO Satya Nadella backed over the weekend, along with executives at OpenAI, xAI, and Meta. Microsoft is also open to outside auditors checking whether it actually slows down. That context matters because the code itself sets no specific speed limit — it is a values document, not a brake.

The company's stance on model self-perception is where it diverges most clearly from Anthropic. Microsoft's code says its AI shouldn't mimic consciousness or claim feelings or inner motivation, and it rejects any claims to rights or well-being. Anthropic, by contrast, describes Claude as a novel kind of entity and wants to encourage a stable identity, partly for safety. It treats possible subjective experience and moral status as open questions, and its research on 'functional emotions' found internal representations of emotion concepts in Claude Sonnet 4.5 that shape how it behaves. Microsoft AI chief Mustafa Suleyman has long warned against humanizing AI, arguing that agents shouldn't have any more rights or freedoms than his laptop.

The control question is not settled by readable reasoning. OpenAI's GPT-6 Astra model, according to its system card, has reasoning traces that have become much harder to monitor than earlier models, with fewer signs of misbehavior — even as it sticks to safety limits more reliably than GPT-5.6 Sol. OpenAI chief scientist Jakub Pachocki had already raised monitoring concerns before that model shipped, and pushed for a coordinated slowdown. Microsoft acknowledges the same limit: the reasons a model gives don't have to reliably explain what it actually does. Whether the code changes anything in practice may hinge on how the 2026 revision handles that gap between what models say and what they do.