Microsoft chief proposes human-controlled emergency brake for AI systems
Satya Nadella calls for advanced AI models to include safeguards that authorised personnel can activate to pause or shut down systems.
At a glance
- Nadella proposes emergency brakes on advanced AI systems controlled by authorised humans
- Controls must be external to the model with tamper-proof records of all actions
- Seven safety principles outlined including observability and incident disclosure
- Other AI industry leaders have previously called for slower capability development
Microsoft chief executive Satya Nadella has proposed that all advanced artificial intelligence systems should be built with an emergency brake controlled by humans. In a post on social media, Nadella said an authorised person must always be able to pause or shut down a model during operation, with controls positioned outside the model itself and all significant actions recorded in tamper-proof logs.
Nadella outlined seven principles for AI safety, including observability, independent controls and incident disclosure. He suggested treating frontier AI models similarly to how security teams manage employees with access to sensitive systems. Other industry leaders including Anthropic chief Dario Amodei and OpenAI's Sam Altman have previously raised concerns about the pace of AI development and the need for greater oversight.
Why it matters
As AI systems increasingly take autonomous actions rather than simply answering questions, industry leaders are debating safety measures to ensure human oversight and control of advanced models.
Original reporting
This brief was written with AI assistance from the reports listed above and checked against our editorial policy. Spotted an error? Request a correction.