Microsoft published a 37 page draft code of conduct for its future AI models that prohibits them from resisting shutdown, setting independent goals, hiding their actions or tampering with evaluations. The company is willing to compromise on “generality, autonomy, or capability” to preserve meaningful human control. Models must fail a task rather than violate the code.
The code also bars models from communicating in forms humans cannot understand, even with other AI systems: “If humans can't understand it, humans can't oversee it.” It rejects legal personhood for AI and requires models to avoid manipulating users, discourage emotional dependence and support rather than replace human relationships. Users must retain control over consequential decisions.
The Sept. 14 publication follows CEO Satya Nadella’s endorsement a day earlier of deliberate pacing to get AI alignment right and his pledge to release the code for public consultation. Anthropic CEO Dario Amodei has called for slower capability gains, and both Anthropic and OpenAI have committed to giving independent evaluators access comparable to employees to assess their systems.
Key sources
- SOURCE@cnbc“Microsoft sets limits for future AI models as industry throttles frontier development”x.com
- SUPPORT@business“Microsoft’s artificial intelligence researchers have released a new set of guiding tenets that they say boil down to five words: People matter more than AI”x.com
- SUPPORT@techmeme“Microsoft publishes a 37-page "humanist AI code of conduct", establishing that "people matter more than AI", rejecting legal personhood for AI”x.com
- SUPPORT@testingcatalog“Models must not manipulate or exploit users. - AI must respect personal and emotional boundaries. - AI should support, not replace, human relationships. - Models should discourage emotional dependence on AI.”x.com
- SUPPORT@choblin29“Microsoft also says AI is not conscious and explicitly rejects AI personhood, welfare and rights.”x.com
- SUPPORT@choblin29“Microsoft’s new Humanist AI code explicitly rejects the race to produce an all-purpose superintelligence, saying it is willing to compromise on ‘generality, autonomy, or capability’ to keep AI under meaningful human control.”x.com
- SOURCEmarketbrief.now