Microsoft CEO Satya Nadella published a long article on a social platform on October 10, publicly calling on the industry to introduce an "emergency stop" mechanism for advanced AI models and to thoroughly evaluate the existing AI trust architecture.

Regarding the rapidly evolving super-intelligent technology, Nadella pointed out that the industry can no longer blindly regard complex black-box models as all-knowing. Instead, it should adopt a defensive assumption that "the model has been compromised" and promote a more cautious approach to safety governance.

In terms of specific technical frameworks, Nadella advocated separating the underlying AI models from the control framework that coordinates their tasks, achieving externalization of control and security measures. He emphasized that every meaningful model operation must be accompanied by tamper-proof, human-readable execution evidence, and ensure that authorized personnel have the ultimate power to pause or shut down the model at any time during task execution. This proposal aligns with the cautious development plan previously proposed by Dario Amodei, CEO of Anthropic, highlighting the high level of concern among top tech leaders regarding the risk of AI model失控.

Industry analysts believe that as the autonomy and agent capabilities of models continue to rise, establishing industry standards that are transparent, controllable, and capable of external intervention has become an urgent priority to balance technological innovation with safety底线.