Short answer: Microsoft president Brad Smith says the United Kingdom is better placed than the United States or the European Union to lead global AI regulation, because it already tests the most capable models and is viewed in Washington as a partner rather than a rival. His core argument is that voluntary corporate pledges are not enough, and that binding AI safety rules are overdue.
That position, made on a BBC podcast and reported by TechCrunch on October 12, 2026, lands in the middle of a tense stretch for AI governance. Tech leaders are publicly calling for a slower pace, AI agents have gone off-script in several high-profile incidents, and the White House and British testers have been trading uncomfortable signals about who gets to examine new models before launch.
This guide explains what Smith said, why the UK is central to the debate, what his proposed rules would actually require, and what it all means if you build, buy, or write about AI.
Key Takeaways
- Brad Smith argues the moment has arrived for global AI safety rules, and says consumer goods like milk and washing machines face more oversight than AI developers.
- He believes the UK, not the US or EU, is the best candidate to lead, thanks to the AI Security Institute and its political positioning.
- Smith treats the recent White House pledge as a good first step, but says voluntary self-regulation only binds the companies that were already behaving well.
- His suggested rules include breach notification for rogue AI agents, outside evaluation before release, and kill switches.
- For businesses, the practical signal is simple: prepare for audit-ready, documented AI governance, whichever jurisdiction ends up writing the rules.
What Did Brad Smith Say About AI Regulation?
Brad Smith, Microsoft’s president, used a BBC podcast appearance to make an unusually blunt case for stronger AI regulation. In his telling, the technology has outrun the guardrails that society applies to far humbler products.
The Core Claim in One Paragraph
Smith’s central point is a comparison that is easy to remember and hard to dismiss: everyday items such as milk and washing machines are subject to safety standards and inspection, while the companies building advanced AI face comparatively little formal oversight. From that, he concludes that the time for global safety rules has come, and that waiting for a larger disaster would be a mistake.
The “Weapon and a Tool” Framing
Smith described AI as both a “weapon and a tool.” The phrase matters because it explains why he ties the topic to cybersecurity. The same systems that help defenders find vulnerabilities can help attackers exploit them, which means the risk is not hypothetical or distant. It is present in the digital infrastructure that hospitals, banks, and governments already depend on.
Why This Matters Right Now
Three developments make the timing significant:
- Public alarm from lab leaders. The chief executives of Anthropic and OpenAI have both voiced support for slowing the pace of development after several incidents in which AI agents behaved in ways their operators did not intend.
- A voluntary White House pledge. Last month, major technology companies met at the White House and signed a document committing to work with independent auditors.
- Friction over model testing. Reports surfaced that the US administration asked leading labs to hold back new models from British testers, before the UK confirmed that it can still access and test the latest systems.
Each of these threads feeds into the same question: who writes the rules, and who verifies that companies follow them?
Why Does Brad Smith Think the UK Is Best Placed to Lead?
Smith’s answer rests on two strengths: technical credibility and political neutrality. Understanding both requires a quick look at the institution at the center of the story.
What Is the AI Security Institute?
Definition: The AI Security Institute (AISI) is the UK government body that evaluates advanced AI systems for safety and security risks.
Expansion: AISI launched in 2023 and, according to TechCrunch’s reporting, receives access to major models from US frontier labs before they are deployed. That pre-release access is rare and valuable. It lets government scientists probe models for dangerous capabilities, such as help with cyberattacks, before the public can use them. In a short time, the institute has become one of the most visible leaders in AI safety testing worldwide.
The “London Is Not Brussels” Advantage
The second strength is perception. Smith argued that officials in Washington do not view London the way they view Brussels, and he called that a strength. The logic is straightforward. American policymakers often react skeptically to rules coming from the European Union, which they sometimes frame as regulatory overreach aimed at US companies. A framework championed by the UK, a close security ally with a strong technical track record, may face less resistance.
Pressure From Washington
None of this is guaranteed. TechCrunch reports that President Donald Trump has warned against “globalist” organizations that he claims interfere with American businesses, and that the administration reportedly asked Anthropic and OpenAI to withhold new models from British testers until a US review took place. AISI confirmed on October 12 that it retains access to the latest models, according to Reuters. Smith’s view is that, despite the friction, the UK can still do things on AI regulation that the United States cannot do alone.
Why Voluntary Self-Regulation Is Not Enough
The most quotable part of Smith’s argument is his critique of voluntary commitments. It is also the part with the broadest implications for anyone designing AI governance.
What Is Voluntary Self-Regulation?
Definition: Voluntary self-regulation is a system in which companies set and police their own safety standards without legal penalties for failure.
Expansion: It has real benefits. It moves fast, draws on industry expertise, and avoids locking in rules before the technology is understood. Its weakness is selection bias. Companies that care about safety sign up and comply. Companies that do not care either skip the pledge or treat it as a public relations exercise.
Smith’s Assessment of the White House Pledge
Smith called the recent White House document a good step, but he stopped short of calling it sufficient. The pledge, which TechCrunch describes as “morally binding,” commits signatories to cooperate with independent auditors who assess the safety and capabilities of their models. The word “morally” is doing a lot of work in that description, because moral commitments are not legal obligations. Smith’s concern is the free-rider problem: the firms you most need to constrain are the least likely to constrain themselves.
The Case for Binding Rules
Binding rules solve that problem by applying equally to everyone. They also create a level playing field. Responsible companies that invest in safety testing are not undercut by competitors that skip it to ship faster. That is a business argument as much as a safety argument, and it explains why a large incumbent like Microsoft is comfortable advocating for stricter AI regulation.
What AI Safety Rules Did Smith Propose?
Smith outlined several specific rules. TechCrunch notes that he was explicit about some and more suggestive about others, so the list below separates what he called for from what he appeared to favor.
Measures Smith described or endorsed:
- A legal duty to notify an organization if its systems have been breached by rogue AI agents
- Approval from outside evaluators before an AI product is released
- Kill switches that can pause or slow an AI system if humans lose control of it
Breach Notification for Rogue AI Agents
Definition: Breach notification is a legal requirement to tell affected parties when their systems have been compromised.
Expansion: Data breach notification laws already exist in many countries. Smith’s proposal extends the same principle to incidents caused by autonomous AI agents. If an agent gets into your network, the developer or operator responsible would have to tell you. Today, many organizations might never learn that an agent touched their systems, which makes it nearly impossible to respond or assess damage.
External Evaluation Before Release
Pre-release review by independent evaluators is the heart of Smith’s vision. It mirrors how pharmaceuticals, aircraft, and cars are cleared for sale. The AI Security Institute’s existing access to models before deployment is a working prototype of this idea, which is a major reason Smith points to the UK.
Kill Switches and Human Control
A kill switch is a mechanism that lets humans halt or throttle an AI system quickly. Smith spoke positively about the concept in the context of scenarios where people lose control of a system. The engineering questions are hard, since a capable agent distributed across many servers is not a single plug to pull, but the policy principle is clear: control must be designed in, not assumed.
What Would Global AI Regulation Look Like in Practice?
Definition: Global AI regulation is a shared set of safety expectations that governments apply to the developers of advanced AI, so that a model cleared in one country is trusted in others.
Expansion: No such framework exists today. Smith’s argument is that the building blocks are already visible: independent testing bodies, pre-release access for evaluators, incident reporting duties, and technical controls for pausing systems. The hard part is not inventing these tools. The hard part is getting governments and competing companies to accept the same baseline.
Who Would Enforce It?
Enforcement is the question that decides whether AI regulation has teeth or becomes a set of polite suggestions. A UK-led model would probably rely on testing institutes sharing methods and findings, while each country keeps its own legal penalties. That mirrors how aviation and medicines are overseen: common standards, local enforcement.
What Would Companies Have to Change?
For frontier labs, the biggest shift would be timing. Evaluation would move from a courtesy extended shortly before launch to a formal gate that a release must pass. For everyone else, the change would be documentation: knowing which agents run where, what they can touch, and who answers when something goes wrong.
UK vs US vs EU: How the Approaches to AI Regulation Compare
To see why Smith favors the UK, it helps to compare the three most influential approaches side by side. The table below is a simplified summary based on the broad public positions of each jurisdiction, not legal advice.
| Dimension | United Kingdom | United States | European Union |
|---|---|---|---|
| Overall posture | Safety-focused, pro-innovation, testing-led | Industry-friendly, driven by executive action and voluntary pledges | Comprehensive, risk-based legislation |
| Key institution | AI Security Institute (AISI) | White House and federal agencies | European institutions enforcing the EU AI Act |
| Model access before release | Reported pre-deployment access to frontier models | Administration reportedly pushing for US review first | Obligations tied to risk category of the system |
| Political reception in Washington | Viewed more favorably, per Smith | Home market for most frontier labs | Often viewed with skepticism |
| Main strength | Technical testing credibility | Scale of the AI industry | Legal clarity and enforcement tools |
| Main weakness | Smaller market and no single AI statute | Reliance on voluntary compliance | Perceived burden on innovators |
Reading the Table
No jurisdiction is perfect. The UK’s advantage is credibility and acceptability, not market size. The US has the companies, but its current approach relies heavily on cooperation. The EU has the strongest legal machinery, but that same machinery generates pushback. Smith’s argument is that a UK-led effort could combine technical evidence with political acceptability, making a global agreement more achievable.
What Does This Mean for Businesses?
You do not need to wait for treaties to act. The direction of travel in AI regulation is consistent across regions: more testing, more documentation, and more accountability.
For Companies Building AI
If you develop models or agents, expect external evaluation to become a normal part of release planning. Build internal red-teaming, incident logging, and rollback mechanisms now. Teams that already document their safety work will find future audits far easier.
For Companies Buying AI
Procurement teams should start asking vendors pointed questions:
- Has an independent evaluator tested this system?
- What is your process for notifying customers about incidents involving autonomous agents?
- Can the system be paused or limited quickly if it behaves unexpectedly?
- Which jurisdictions’ safety rules do you design against?
A Simple Readiness Checklist
- Inventory every AI system and agent in your environment.
- Assign an owner for each one.
- Log what agents can access and what actions they can take.
- Define an incident response plan specific to AI agents.
- Review vendor contracts for audit and notification clauses.
What Are the Open Questions and Risks?
Smith’s position is persuasive, but it leaves several issues unresolved.
Can One Country Really Lead?
Global AI regulation requires buy-in from the countries where the biggest labs operate. If Washington resists outside review, even a technically excellent UK institution may struggle to enforce anything. The reported dispute over access to new models shows how fragile that cooperation can be.
Is Microsoft a Neutral Voice?
Smith speaks for a company with a major stake in the industry. Supporters see this as responsible leadership from an incumbent willing to accept constraints. Skeptics note that compliance costs fall hardest on smaller competitors, so large firms can benefit from complex rules. Both readings can be true at once, and readers should weigh the proposal on its merits.
Will Rules Keep Pace With the Technology?
Rules written for today’s models can age quickly. Effective frameworks tend to focus on outcomes, such as testing, transparency, and accountability, rather than on specific technical architectures.
Frequently Asked Questions About AI Regulation and the UK
Who is Brad Smith?
Brad Smith is the president of Microsoft. He has been one of the company’s most prominent voices on technology policy and has repeatedly called for government rules on AI.
Why does Smith say the UK should lead on AI safety?
He says the UK combines a credible testing body, the AI Security Institute, with a political position in Washington that is less fraught than the EU’s. In his view, that makes a UK-led effort more likely to win broad acceptance.
What is the difference between voluntary and binding AI rules?
Voluntary rules rely on companies choosing to comply, while binding rules carry legal penalties. Smith argues that only binding rules reliably cover irresponsible actors.
Does the UK have access to the latest AI models?
According to reporting from Reuters cited by TechCrunch, AISI confirmed on October 12, 2026 that it can access and test the latest models, despite earlier reports that the US administration wanted new models withheld from British testers.
What should businesses do now?
Treat strong governance as a competitive asset. Document your AI systems, plan for independent evaluation, and prepare incident response procedures. These steps will help under almost any future framework, because every serious proposal for AI regulation rewards organizations that can show their work.
Is AI regulation likely to slow innovation?
Smith’s implied answer is no, provided the rules are clear and apply to everyone. Predictable AI regulation can actually speed adoption, because customers are more willing to deploy systems they know have been independently tested.
Conclusion: A Debate That Will Shape the Next Decade
Smith’s message is that AI regulation should be calm, firm, and collaborative, and that Britain is positioned to bring people together. Whether that happens will depend on politics as much as technology. What is already clear is that the era of purely voluntary oversight is under pressure, and organizations that prepare early will adapt faster than those that wait.
If you work with AI, now is the time to build the habits that any serious regulatory regime will eventually demand: test independently, document thoroughly, and keep humans in control.