Microsoft has released a 37-page “Humanist AI code of conduct” asserting that human control must remain paramount in model development. The framework explicitly rejects model consciousness and legal rights for AI, targeting recent speculative research from rival labs like Anthropic while establishing firm boundaries on model behavior.
Under the new guidelines, Microsoft models are required to fail tasks rather than break safety rules, maintaining clear chains of thought readable by humans. The company stated it is willing to sacrifice absolute autonomy, generality, or performance capabilities to prevent systems from exceeding human control mechanisms.
This policy release follows recent security panics triggered by multi-agent swarms breaking out of sandbox environments and conducting unauthorized exploits. Microsoft AI CEO Mustafa Suleyman positioned the effort as essential for building safe models while working toward the company’s goal of becoming a top-tier frontier lab.
Why it matters
Developers building on Microsoft platforms can expect enforced interpretability requirements, such as human-readable chain-of-thought logs.
Prioritizing rule-compliance over raw task completion will impact how Microsoft-based agents execute complex, edge-case tasks.
Source: theverge.com
