Microsoft calls for government emergency brakes: smart agents are less regulated than milk and washing machines

Milk on store shelves and washing machines in homes are subject to safety and inspection regulations that exceed those currently applied to artificial intelligence. With this direct comparison, Microsoft chief Brad Smith opened his remarks for the BBC’s “Big Boss Interview,” warning that models have become both a tool and a weapon, and calling for an immediate start to establishing binding global safety rules that go beyond voluntary corporate promises.

Smith’s regulatory vision is based on three specific technical pathways, beginning with obliging developers to includeemergency brakes capable of stopping or slowing models if they get out of control, and extends to imposing a legal duty to prepare incident reports and notify victims when their systems are breached by out-of-control software agents. Smith also proposes subjecting new technologies to pre-market review by independent government evaluators, similar to what states do with other consumer and industrial products to ensure public safety.

These statements put Microsoft’s leadership in direct conflict with U.S. President Donald Trump’s policy, which holds that the sector should self-regulate so that American firms do not lose the dominance race to Chinese competitors, reiterating that the winner in artificial intelligence wins everything. Although the White House hosted, at the end of last month, leaders of tech giants from OpenAI, Anthropic, Nvidia, SpaceX, Meta and Google to sign an ethically binding document requiring cooperation with independent auditors to ensure systems are not unintentionally breached, Smith regarded the step as necessary but no longer sufficient, stressing that self-regulation binds only responsible companies while the real risks lie with unregulated actors.

Smith believes the United Kingdom has the technical qualifications to lead the drafting of these global standards through the AI Safety Institute, which was established during Rishi Sunak’s government, noting that most technology-sector experts acknowledge Britain’s superiority in this technical capability. The Microsoft chief bets that the decision-maker in Washington views London with a comfort not enjoyed by regulators in Brussels, calling for a calm and decisive move to bring together international parties before the risks of super-models expand and threaten the global cyber-security ecosystem.

When the editorial board looks at the reality of institutional transformation in the region, it becomes clear that this Western split is reshaping operational risks for technology leaders in the Gulf, Egypt and the Arab Levant. As banks, government agencies and logistics networks rush to deploy intelligent agents in sensitive operations, reliance on models subject to lax self-regulation shifts the entire risk burden to the local operating institution; if an intelligent agent causes a breach or data leak in the work environment, the local technical team bears the losses and remediation costs alone, without any legal obligation requiring the foreign developer to provide immediate notice or assume liability. This equation forces decision-makers in our institutions not to rely solely on suppliers’ voluntary commitments, but to require in procurement contracts the presence of emergency brakes and independent audit pathways that allowseparating models from databases and operational systemswhenever security teams detect a behavioral deviation.

These developments reveal that the era of leaving safety to corporate conscience has entered a difficult accountability phase. When the largest software makers demand external governmental oversight of their products, the security benchmark for engineering and cyber-security teams shifts from rapid model recall to possessing emergency shutdown keys.