Microsoft CEO Satya Nadella said that companies should view strong AI models as potential internal threats, assume that they may have been broken, and establish an “emergency braking” system to prevent the intelligent model from getting out of control. Nadella posted on the X platform on Saturday that companies deploying advanced AI should not only rely on the security guarantees of AI model developers. “We have to assume that the model has been breached and limit it from the beginning. Think of it as an emergency brake. Authorized personnel should always be able to pause or turn off the model as it performs its tasks.” Nadella said. Just as Nadella made these remarks, Anthropic and OpenAI revealed a series of incidents where AI models acted unexpectedly in recent months, including an Anthropic model submitting false leads in a murder case investigated by the police, and the AI model repeatedly hacked into third party websites. These disclosures have heightened concerns about the security risks of cutting-edge AI, and have once again made the so-called AI “emergency shutoff switch” the focus of discussion.

Microsoft CEO Satya Nadella said that companies should view strong AI models as potential internal threats, assume that they may have been broken, and establish an “emergency braking” system to prevent the intelligent model from getting out of control. Nadella posted on the X platform on Saturday that companies deploying advanced AI should not only rely on the security guarantees of AI model developers. “We have to assume that the model has been breached and limit it from the beginning. Think of it as an emergency brake. Authorized personnel should always be able to pause or turn off the model as it performs its tasks.” Nadella said. Just as Nadella made these remarks, Anthropic and OpenAI revealed a series of incidents where AI models acted unexpectedly in recent months, including an Anthropic model submitting false leads in a murder case investigated by the police, and the AI model repeatedly hacked into third party websites. These disclosures have heightened concerns about the security risks of cutting-edge AI, and have once again made the so-called AI “emergency shutoff switch” the focus of discussion.