Nadella puts Microsoft's MAI code of conduct up for public consultation and backs 'deliberate pacing'
Microsoft chief executive Satya Nadella wrote on X on Sunday that any pursuit of superintelligence must rest on AI that helps humanity and stays under human control, that the company welcomes the 'deliberate pacing' needed to make alignment the design goal, and that the code of conduct behind its own MAI models would be published the next day for public consultation. The post is Microsoft's answer to Dario Amodei's call to pace the frontier; the text of the code had not appeared when the reports were written.
Artificial Intelligence··Midday
'Not worth pursuing' without human control
Satya Nadella, Microsoft's chairman and chief executive, wrote on X on Sunday 13 September that any pursuit of superintelligence has to be grounded in the principle that AI which is not helping humanity and under human control is not worth pursuing. In that context, he wrote, Microsoft welcomes the research, focus and 'deliberate pacing' required to make alignment the design goal, ideas such as 'embedded evaluators', and the broader effort to build mechanisms that make this more than talk. Such work, in Nadella's view, cannot be controlled by 'a handful of entities' and needs broad representation across the ecosystem, countries and fields, academia included. The post is among the company responses to 'We Must Pace the Frontier', the essay in which Anthropic chief executive Dario Amodei argued that the rate of capability improvement must slow so that risk prevention can keep up.[1], [2]
Open and closed models, and firms that keep their own learning loops
Nadella also called for the benefits of AI to be accelerated and spread broadly across countries, communities and companies, which he said requires a 'frontier ecosystem' in which both closed and open-source models can thrive. For firms he set out a further condition: each organisation should keep full control over its 'unique and tacit knowledge', be able to build its own continuous learning loop without depending on any single model provider, and embed that knowledge into models and weights it controls. He summarised Microsoft's own approach in three parts: broad access and choice at every layer of the AI stack, enterprise control of learning loops and models, and the code of conduct underlying the company's first-party MAI models.[1], [2]
The code of conduct announced for 14 September
The third part is the concrete commitment: Nadella said the code of conduct behind Microsoft's own MAI models would be published 'tomorrow', on 14 September, for public consultation. MAI is the name of Microsoft's first-party models, the ones the company develops itself rather than licensing from a partner. When Fox Business and Unite.AI published their accounts the text of the code had not yet appeared, and what it asks of the models, how long the consultation runs and who may take part were not stated. In the essay the post answers, Amodei had proposed embedded third-party evaluators with ongoing, employee-like access to frontier labs, coordination among frontier companies in democratic countries on shared safety standards and limits on unchecked progress, and global coordination, naming METR as an example evaluator.[2], [1]
