On October 11, Microsoft CEO Nadella reminded enterprises that powerful AI models should be regarded as potential internal threats, assuming they may have been compromised, and an 'emergency brake' mechanism should be established to prevent autonomous agent models from getting out of control. Nadella emphasized in a social media post on Saturday that restrictions should be imposed on models from the outset, and authorized personnel should be able to pause or shut down the models at any time. His safety recommendations include: not relying on a single model for critical decision-making, maintaining tamper-proof records of agent behavior, and subjecting systems to independent audits. Additionally, he called on enterprises to disclose major failures or security vulnerabilities, sharing the causes and details of incidents to help other enterprises enhance their security protections.