Brad Smith, president of Microsoft, called for urgent global rules to regulate artificial intelligence, warning that companies developing these systems face less oversight than consumer products such as milk and washing machines. Smith said on a BBC podcast that artificial intelligence can be both a “weapon and a tool,” and that it threatens cybersecurity, making the adoption of global safety rules urgent.
Smith believes that the United Kingdom, rather than the United States or the European Union, is best positioned to lead this effort. This assessment is based on the role of the UK AI Safety Institute (AISI), which was launched in 2023 and enables leading US laboratories to access large models before they are released, with the aim of testing them and assessing their risks.
Model testing under political pressure
The institute’s role came under threat in recent weeks after US President Donald Trump warned against organizations he described as “global” and which intervene in the interests of American companies. The article said that Trump had reportedly asked Anthropic and OpenAI to prevent British testers from accessing new models, but the institute confirmed on Monday that it could access and test the latest models.
Smith believes London has regulatory space that is unavailable to Washington. He said policymakers in the US capital do not view the United Kingdom and London in the same way they view Brussels, considering this a strength when attempting to bring parties together around shared rules.
What changes in practice?
Smith proposes moving from voluntary commitments to enforceable legal obligations. Among the measures he mentioned was requiring organizations to notify an authority when their systems are breached by artificial intelligence agents that have gone out of control, along with requiring AI laboratories to obtain approval from independent assessors before launching products.
He also spoke positively about kill switches that would allow artificial intelligence systems to be slowed or suspended if humans lost control of them. These calls follow similar demands from leaders at Anthropic and OpenAI to slow the development of artificial intelligence, following incidents in which software agents behaved in an uncontrolled manner.
Limits of voluntary regulation
Leaders of companies including OpenAI, Anthropic, Nvidia, AMD, SpaceX, Meta and Google had signed a “morally binding” document at the White House pledging to cooperate with independent auditors to assess the safety and capabilities of models. Smith described the document as a good step, but said it did not reach the level of regulation required, because voluntary compliance may be limited to responsible companies, while the greatest risks lie with entities that are less committed.
Smith’s statements illustrate a growing trend within the artificial intelligence sector toward rules requiring testing, reporting and control mechanisms before launch, but they do not in themselves constitute government policy or an international agreement. The open question remains whether the United Kingdom will retain the ability to test the latest models, and whether cooperation between laboratories and regulators will become binding legal obligations.