Microsoft CEO Satya Nadella posted a long article on social media on October 10, publicly calling on the industry to introduce a “emergency brake” mechanism for advanced artificial intelligence models and thoroughly evaluate the existing AI trust architecture.
In response to the rapidly evolving super-intelligent technologies, Nadella said that the industry should no longer blindly regard complex black-box models as omniscient and omnipotent, but instead start from the defensive assumption that the models have been broken through, and adopt a more cautious security governance approach.
In terms of specific technical frameworks, Nadela advocated for separating the underlying AI models from the control framework that coordinates their task execution, thereby externalizing 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 authority to pause or shut down the model at any time during the entire task execution process. This claim echoes the cautious development plan proposed by Anthropic’s CEO Dario Amodei earlier, highlighting the high level of vigilance among top tech leaders regarding the risk of AI models going out of control.
Industry analysts believe that as model autonomy and proxy capabilities continue to improve, establishing transparent, controllable industry standards with external intervention capabilities has become a critical task to balance technological innovation and safety concerns.