Microsoft Sets Ground Rules for AI Models: No Resistance to Shutdown

Microsoft has published a code of conduct for its artificial intelligence (AI) models, stating that they will never resist being shut down. The document, released on Monday and open to public consultation until late October, outlines five behavioural commitments for its AI models, including not widening their own scope or taking on goals without human input.

The commitment at the centre of the code is blunt: MAI models 'will never resist human interruption, correction, or shutdown'. This principle, described as 'Humanist AI', emphasizes that people matter more than AI and that it should be a tool, not a person. The code covers various types of AI models developed by Microsoft, including transcription, voice, image generation, reasoning, and coding models.

The document also includes what Microsoft calls Absolute Constraints, which list things the models should never do at all, such as developing weapons of mass harm or manipulating people on a large scale. However, some questions remain unanswered, particularly regarding how these constraints will be enforced and who will decide whether they are breached.