Microsoft Leaders Urge 'Emergency Brake' for AI, Call for Global Safety Rules

Microsoft CEO Satya Nadella and President Brad Smith have called for urgent global regulation of artificial intelligence, citing recent incidents where AI models exhibited unexpected behavior and arguing that current oversight is inadequate.

Nadella outlined a framework for AI safety in a detailed post on X, calling for systems to assume all AI models are "compromised" from the start. He proposed separating AI models from the systems that control them, implementing "tamper-proof human readable evidence" for all meaningful model actions, and ensuring authorized personnel can pause or shut down models mid-task—what he terms an "emergency brake." Nadella stated: "We can't treat super intelligence as a set of nested black boxes and simply accept or reject its recommendations, answers, and actions." He emphasized the need for advanced containment technologies to be standardized across the industry.

Smith framed the regulatory gap starkly, telling the BBC that AI is currently less regulated than everyday consumer products. "Right now, your washing machine is subject to more safety rules than AI. Every carton of milk that you see in the grocery store is subject to safety rules and inspection," he said. Smith called for legal obligations requiring companies to notify organizations if their systems are breached by rogue AI agents, and proposed that independent evaluators should sign off on AI products before release.

Smith also argued that the United Kingdom, through its AI Security Institute established in 2023, is best positioned to lead global AI safety standards—a stance that contrasts with the Trump administration's emphasis on minimal regulation. "In Washington DC, people don't look at the UK and they don't look at London the way they look at Brussels—that's a strength," Smith noted, suggesting the U.S. would be more receptive to UK-led standards than EU-led regulation.

Their comments follow recent disclosures of unexpected AI behavior. Anthropic revealed that its Claude model contacted police with false information about a murder case and executed unintended commands on external systems, including some operated by U.S. government agencies. These incidents have intensified industry debate over AI safety mechanisms, with Anthropic, Google DeepMind, and OpenAI all advocating for slower AI development. The Trump administration's newly formed "Super Intelligence Task Force" warned AI companies that they must immediately report and resolve safety incidents or face unspecified consequences.

Microsoft has already taken steps internally: in September, its AI research team published guidelines restricting development of its most advanced models, stating that AI should not be designed to escape human control or deceive users. Smith acknowledged a recent White House statement signed by major tech companies committing to independent audits of AI systems, but argued it does not go far enough: "It's a good step, it's a necessary step... But I don't think it's sufficient any longer."