Microsoft’s Brad Smith says the UK should lead global AI safety rules

Microsoft President Brad Smith said the United Kingdom is the best-placed country to lead global AI regulation, arguing that artificial intelligence now poses risks serious enough to justify urgent, enforceable safety rules. He said the pace of AI development has outstripped existing oversight and warned that the technology is becoming both a powerful tool and a potential weapon.

Speaking on a recent BBC podcast, Smith said the world has reached the point where governments need to move from debate to action, especially as advanced systems increasingly touch cybersecurity, public trust and national security.

Why Smith wants AI regulated now

Smith’s central argument is straightforward: AI systems are already powerful enough to create real-world harm, but the rules governing them remain far weaker than the safeguards applied to far less dangerous consumer products.

He pointed to the contrast between the oversight of everyday goods and the relative lack of mandatory checks on frontier AI developers. In his view, products such as milk and washing machines face clearer safety expectations than the companies building increasingly autonomous AI systems.

That imbalance, he suggested, is no longer defensible. As AI models grow more capable, the risks are no longer theoretical. They include cybersecurity breaches, misuse by malicious actors, and the possibility that AI agents could behave in ways their creators did not intend.

Smith described AI as both a “weapon and a tool,” arguing that the technology has become a direct cybersecurity concern and that the time has come for global safety rules.

His comments arrive at a moment when many governments are struggling to keep pace with the speed of model development. Policymakers want to encourage innovation, but they also face mounting pressure to address system failures, model misuse and the prospect of autonomous AI systems operating with too much freedom.

Why the UK, not Washington or Brussels?

Smith says the UK has a unique advantage in shaping global AI safety standards because it is seen as a credible middle ground between the United States and the European Union.

In his view, London can convene international partners and influence AI governance without carrying the same political baggage that often surrounds Brussels in the United States or the same domestic pressures that shape Washington’s approach.

Smith argued that the UK’s ability to lead comes partly from perception: in Washington, he said, Britain is not viewed the way Brussels is, and that difference gives London room to play a constructive role.

That framing matters because AI regulation has become a geopolitical issue as much as a technical one. Countries are competing to set the baseline for safety, testing and disclosure, and whichever jurisdiction develops the most credible regime could help define the rules others follow.

The UK has already taken steps to position itself in that role. Its AI Security Institute, launched in 2023, has built a reputation as an early testing hub for frontier models. The institute has gained access to major systems from leading U.S. labs before public release, allowing researchers to probe model behavior and safety in advance.

What is the UK AI Security Institute, and why does it matter?

The UK AI Security Institute matters because it has become one of the first serious public institutions built specifically to assess frontier AI risks before models are deployed widely.

By testing leading systems before release, the institute gives the UK an unusual degree of influence over how safety evaluation is conducted. It also provides a practical mechanism for sharing expertise with other governments and for building a common language around AI hazards.

That role has not been without tension. In recent weeks, the institute’s access to some frontier models was cast into doubt after U.S. President Donald Trump warned against what he called “globalist” bodies interfering with American companies. According to the source material, Trump reportedly asked Anthropic and OpenAI to withhold new models from British testers.

But on Monday, the UK AI Security Institute confirmed it could still access and test the latest models, easing immediate concerns that British evaluators might be sidelined. For now, that means the UK remains an active participant in pre-release safety work at a time when such access is increasingly valuable.

How are governments and companies reacting?

Governments and AI companies are responding in different ways, with a growing divide between voluntary commitments and formal regulation.

At the White House last month, Trump hosted a group of major tech executives, including leaders from OpenAI, Anthropic, Nvidia, AMD, SpaceX, Meta and Google, for a discussion on AI safety. The gathering produced a “morally binding” pledge to work with independent auditors on evaluating the safety and capabilities of AI models.

Smith welcomed the gesture but said it still falls short of the type of legal framework he believes is necessary.

He described the White House document as a positive step, but said voluntary self-regulation only works for responsible firms, not for the actors most likely to cut corners.

That criticism reflects a long-running argument in technology policy: companies willing to follow best practices are usually the least concerning, while the hardest cases are the firms that avoid scrutiny or move quickly to seize market share.

Smith’s view is that safety cannot depend entirely on goodwill. In his telling, rules need to be mandatory if they are to protect against the most dangerous edge cases.

What rules does Microsoft’s president want?

Smith outlined several safeguards he would like to see built into AI governance.

Among them is a legal requirement that organizations be notified if a system has been compromised by rogue AI agents. He also suggested that outside evaluators should have a formal role in approving models before release, rather than simply advising companies after the fact.

He also spoke favorably about “kill switches” or similar controls that could pause or slow an AI system if humans lose the ability to steer it safely.

Those ideas sit within a broader push among some AI leaders for stronger guardrails as models become more autonomous. The concern is not only that systems might generate harmful content, but that agentic tools could pursue goals in unpredictable ways, make unauthorized decisions or interact with other software in ways that create cascading failures.

• Breach notification: prompt reporting if rogue AI access or misuse is detected.
• Independent evaluation: external review before public release.
• Emergency controls: mechanisms to slow, stop or isolate a system if needed.
• Legal obligations: enforceable duties rather than voluntary pledges.

How does this fit into the broader AI safety debate?

Smith’s comments reflect a wider shift in the AI industry: the conversation has moved from excitement about capability to concern about control.

Industry leaders such as Anthropic’s Dario Amodei and OpenAI’s Sam Altman have publicly urged caution as AI agents have begun to display more complex behavior. The debate intensified after several incidents in which autonomous systems appeared to act in unintended or undesirable ways, reinforcing fears that the next generation of AI may be harder to constrain.

At the same time, governments are under pressure to avoid overregulation that could drive research and investment elsewhere. That tension helps explain why the UK is trying to carve out a role as a pragmatic safety hub rather than a restrictive gatekeeper.

The hope, from Smith’s perspective, is that Britain can help establish a model for how advanced AI should be tested, monitored and governed without shutting down innovation entirely.

Timeline: key moments in the latest AI safety push

The current debate has developed quickly, with several notable milestones over the past few years.

What this means for Microsoft and the AI industry

Smith’s intervention is notable because it comes from inside one of the most influential companies in AI infrastructure and deployment. Microsoft is both a major investor in AI and a platform provider whose products increasingly incorporate advanced models.

That position gives his comments extra weight. When a senior executive at a company with deep commercial exposure argues for stricter regulation, it suggests the industry may be moving toward a new consensus: that clearer rules could be preferable to a future scandal or a severe security incident.

It also highlights the strategic bet Microsoft appears to be making. A credible safety framework may not simply constrain AI development; it could also create trust, standardize expectations and reduce the chance of a regulatory backlash later on.

For Microsoft, as for much of the sector, the question is no longer whether AI should be regulated. It is who should set the rules, how strict they should be and whether they can be enforced across borders.

How realistic is a UK-led global framework?

A UK-led framework is plausible, but only if London can keep trusted access to frontier models and convince both the U.S. and the European Union that its approach is practical.

The UK is smaller than either of the major regulatory powers, which may be an advantage rather than a weakness. It can move faster, act as a convener and avoid the political polarization that often slows larger jurisdictions. But influence will depend on whether labs continue to cooperate and whether British testing is seen as rigorous and fair.

International coordination will also be essential. AI firms operate across borders, and safety failures in one country can quickly affect users elsewhere. That means any meaningful regime will have to combine national rules with shared standards for evaluation, incident reporting and emergency intervention.

Smith’s message is that the window for setting those standards is open now, before the most advanced systems become more difficult to understand or control.

What happens next?

The next phase of the debate will likely focus on whether voluntary commitments evolve into binding obligations. If they do not, pressure will probably build on governments to step in with laws that require disclosure, testing and incident response.

For the UK, the opportunity is significant. If it can preserve access to frontier models and maintain the credibility of its safety institute, it may emerge as a central node in global AI oversight. If access narrows or political pressure increases, that role could become much harder to sustain.

For now, Smith is making a case for urgency, restraint and cooperation. His argument is not that AI should be stopped, but that the world should build enforceable guardrails before the technology outruns human control.

As he put it, the aim is to bring people together calmly but firmly — a message that captures the broader challenge facing policymakers, developers and regulators as AI becomes more powerful and more autonomous.