Microsoft CEO Satya Nadella has outlined a call for improved safety measures in artificial intelligence development. In a post on X, Nadella stressed the importance of reassessing the trust framework around AI systems, cautioning against treating advanced AI as opaque black boxes whose outputs are accepted without scrutiny. He proposed separating AI models from the systems that manage their operations and implementing external controls and safeguards.
Nadella highlighted the necessity of documenting every significant AI action with tamper-proof, human-readable records. He also insisted that authorized individuals must retain the ability to pause or shut down AI models during their tasks, likening this capability to an "emergency brake" designed to contain potential compromises from the outset.
These remarks come amid growing concerns in the AI industry about losing control over increasingly complex models. They follow similar cautionary approaches, such as those advocated by Anthropic CEO Dario Amodei, who recently outlined plans for more cautious AI development. Nadella’s emphasis on transparency and control reflects a broader industry push to address AI safety as models become more powerful and integrated into critical applications.