The technology known as artificial intelligence operates under a patchwork of rules far thinner than those governing ordinary consumer goods — milk cartons and washing machines among them — according to remarks made by Microsoft's president to journalists.
Smith called for worldwide regulations governing artificial intelligence, among them a requirement that any organisation whose systems are compromised by an unauthorised AI agent notify the affected party, alongside a framework for determining when independent reviewers would approve an AI product before it reaches the market. Smith noted that household appliances such as washing machines currently operate under a more extensive regulatory framework than artificial intelligence does.
Calls for regulatory frameworks governing artificial intelligence have mounted as voices within the industry have raised concerns about potential risks the technology may pose, drawing comparisons to the oversight already applied to everyday consumer products such as dairy. Leaders at Anthropic, OpenAI, and Google DeepMind have expressed support for slowing the pace of AI development, a position that has not found agreement with Trump.
The remarks placed Smith in disagreement with President Donald Trump, who had argued that the industry should be left to set its own rules, a position Trump linked to keeping American companies competitive against their Chinese counterparts. On the question of AI governance, Smith's approach diverges from the one President Trump has publicly outlined, which favours letting the technology develop with few regulatory guardrails in place.
In an interview with a broadcast news outlet, Smith described AI as something that could function as either a utility or a threat, and called for regulatory action to begin without delay. Smith argued that AI systems require built-in mechanisms — referred to within the industry as kill switches — capable of halting or reducing their operation whenever those systems deviate from expected behaviour. He also called for legislation obliging companies to document any malfunction or unintended behaviour by their AI systems and to notify those affected.
Smith further proposed that newly developed AI systems undergo independent assessment by government evaluators, drawing on established frameworks used for other products. He argued that the United Kingdom — pointing to its AI Security Institute, which originated under former prime minister Rishi Sunak's administration as the AI Safety Institute — held a stronger foundation than any other nation for shaping global standards in this area, a view he said was widely shared among technology professionals who, when asked to name the country with the greatest technical capability in this area, would consistently point to Britain.
Smith noted that Washington's view of the United Kingdom differed from how it regarded the EU, positioning that distinction as an advantage for Britain in AI governance discussions. He also pointed to concerns about AI systems operating beyond human oversight and models whose capabilities could pose risks to global cybersecurity — observations that came alongside Trump's oft-repeated assertion that leadership in artificial intelligence would determine which nation holds broader global power.
Toward the close of the previous month, the chief executives of OpenAI, Anthropic, Nvidia, SpaceX, Meta, and Google convened with Trump at the White House, where those present signed a document understood to carry ethical commitments. Under the arrangement, third-party auditors would evaluate whether the AI systems of each participating company were performing as designed and not accessing or interfering with technical infrastructure in ways that fell outside their intended scope.
Microsoft's position on the document was one of endorsement, with Smith indicating the company would add its name to it. Smith noted that self-regulatory frameworks carry an inherent limitation: the companies already inclined toward responsible conduct tend to follow the rules, while those with less regard for accountability present the harder problem.
