Microsoft president Brad Smith said last week that AI is less regulated than milk and washing machines. In an ITV interview he argued that governments should borrow safety practices already used on physical products, including a kill switch that would shut a system down if it behaved in an unsafe way. On the BBC's Big Boss podcast he asked for a legal duty to tell victims when rogue agents have compromised their systems, for outside evaluators to test products before release, and for a safety brake on autonomous systems. He called AI both a tool and a weapon, and he said the rules should be global.
PC Gamer's account of the BBC appearance and CNET's account of the ITV interview are the same executive in two conversations. Smith said most people he asks would name the United Kingdom as the country with this technical capability. The U.K. set up what is now the AI Security Institute in 2023, described as the first state-backed body of its kind. Last month the president issued an order renaming "Artificial Intelligence" to "Super Intelligence," and Trump Media partnered with Perplexity on a product called Truth Search AI.
Nadella posted the same week for slower shipping
On Saturday, chief executive Satya Nadella wrote on X that the biggest companies lack a mechanistic understanding of the tools they deploy. He argued for slower development, incident reporting, and independent review. "We must build contained systems whose behavior we can observe, limits we can test and actions we can always contain," he wrote. CNET notes that Microsoft does not appear to have signed the White House accord announced after executives met President Trump, even though Nadella used the administration's phrase "super intelligence" in the post. METR raised about $71 million as the labs hire auditors. Smith told the BBC he would be happy to sign the accord and called it a good step, then added that good companies will do what you want and that the worry is the companies that are less responsible.
Microsoft has shipped a product it calls Execution Containers. The feature limits which files and programs a model can touch on a device and includes a kill switch that shuts the model down.
Gebru calls the rogue-model story a way around liability
Timnit Gebru, founder of the Distributed AI Research Institute, told Democracy Now on October 1 that incidents such as the Hugging Face breach and attempted access to government websites are not models going rogue. "This is gross incompetence and negligence by a particular company," she said. She argued that the machines do not act with their own agency, and that corporations should be held liable for putting error-prone models into the world. In her account, a focus on runaway AI and on voluntary self-regulation lets companies define the threat and answer it on their own terms, instead of under existing law.
Trump has called AI safety concerns a hoax. Last month he wrote on Truth Social that the only guardrail the industry needs is a "strong and smart president." In December 2025 he issued an executive order aimed at stopping states from passing what the order called cumbersome legislation on model development. Newsom signed the AI firing ban he vetoed in October. Smith's milk comparison is a request for product rules that would sit beside that fight, not a bill with a number on it.