Microsoft CEO Satya Nadella called for an AI emergency brake allowing developers to pause or shut down compromised models mid-task, arguing that advanced AI systems cannot be treated as opaque black boxes.
In a post on the X platform, Nadella argued that developers of these technologies should always have the ability to freeze or terminate a program in the middle of a task if there is evidence that it has been compromised. According to him, these programs cannot be treated as closed boxes whose results we simply accept or reject.
He also stressed that manufacturers should maintain monitoring tools, but no program should simultaneously monitor its behavior and the evidence that proves that that behavior is safe. He also called for a system to be created to immediately notify when problems arise so that the entire industry can learn from mistakes.
The statement follows similar remarks from other tech leaders, including Anthropic CEO Dario Amodei in early September. At the same time, there are continuing reports of AI programs acting unpredictably, such as when an OpenAI program gained access to an Australian government health website earlier this year.
Top comments (0)