Microsoft Chief Warns AI Is Less Regulated Than Milk and Washing Machines- Artificial intelligence is developing faster than the rules designed to keep it in check. And Brad Smith, president of Microsoft, thinks that is a problem governments can no longer afford to ignore.
His comparison is striking: milk sold in supermarkets and washing machines used in homes face safety requirements that AI systems do not. As the technology becomes more powerful and finds its way into everyday life, Smith believes it is time to ask a difficult question: why are there still so few clear rules governing AI?
Speaking to the BBC’s Big Boss Interview, Smith called artificial intelligence both a “tool and a weapon”. He argued that governments need to establish safety standards before a serious incident forces them to act.
His comments come amid a growing disagreement over how AI should be managed, with some technology leaders calling for stronger safeguards while US President Donald Trump pushes for fewer restrictions to help American companies stay ahead of China.
AI Companies Cannot Be Left to Set All the Rules
For years, much of the debate around AI regulation has centred on how to encourage innovation without creating unnecessary obstacles for developers. But as AI systems become capable of carrying out more complicated tasks with less human involvement, concerns about their potential misuse are becoming harder to dismiss.
Smith wants governments to establish clear legal responsibilities for the companies developing these systems. In his view, safety should not depend entirely on whether a company chooses to introduce safeguards voluntarily.
One of his proposals is to require certain AI products to undergo independent assessments before they are released. External experts could examine their capabilities, identify weaknesses and determine whether additional precautions are necessary.
That would represent a significant change in how some AI products reach the market. It would also raise questions about which systems need independent approval and how regulators can keep pace with technology that changes so quickly.
Smith believes there is already enough experience from other industries to help governments get started.
What Happens When an AI System Goes Wrong?
Smith is also calling for emergency controls that allow AI systems to be stopped if they begin behaving dangerously.
Often referred to as kill switches, these mechanisms could help operators intervene when an AI agent acts outside its intended boundaries. The idea is particularly relevant as businesses increasingly explore AI agents that can access software, process information and perform tasks with limited supervision.
A system that makes a mistake while completing a routine task may cause little harm. One with access to sensitive data or important business operations could create a much bigger problem.
An emergency shutdown mechanism would not prevent every failure, but it could help limit the damage. Making such controls effective, however, would require proper testing and clear procedures for deciding when to activate them.
Smith also wants companies to face a legal obligation to report serious AI incidents and inform people affected by rogue AI agents.
That requirement could make a difference when users need to know whether an AI system has acted without authorisation or exposed them to harm. It could also help regulators identify patterns of failure instead of relying on scattered reports from individual companies.
Trump’s AI Strategy Takes a Different Direction
Smith’s call for stronger safeguards puts him at odds with Trump’s approach to AI development.
The US president has repeatedly stressed the importance of winning the global AI race, particularly against China. His administration has favoured a less restrictive approach, arguing that American companies need room to innovate and compete.
There is a real concern behind that argument. Regulations that are too broad or complicated could increase costs, discourage smaller developers and slow the introduction of useful technology.
But moving too quickly carries its own risks. If companies release increasingly capable systems without adequate testing or accountability, the consequences could extend well beyond the technology industry.
The challenge is finding a middle ground: rules strong enough to address genuine dangers, but flexible enough to accommodate new developments.
Why the UK Is Part of the Conversation
Smith believes the UK is well placed to help shape international AI safety standards. He pointed to the country’s AI Security Institute, established under former prime minister Rishi Sunak’s government as the AI Safety Institute.
The institute’s technical expertise could help inform how advanced AI systems are tested and assessed. The UK could also help bring governments and technology companies together to agree on common approaches to safety.
International cooperation matters because AI does not stop at national borders. A system developed in one country can be deployed in another, making it difficult for individual governments to address every risk on their own.
Common standards could make it easier to assess dangerous capabilities, share information about serious incidents and establish clearer expectations for developers.
The Question Is No Longer Whether AI Needs Oversight
Artificial intelligence has enormous potential, from helping researchers make discoveries to improving services and automating time-consuming work. The aim of regulation should not be to shut down that progress.
But innovation also brings responsibility. If AI systems are going to make decisions, handle sensitive information and act on behalf of people, there must be a clear way to hold their developers and operators accountable when things go wrong.
Smith’s argument is straightforward: safety measures should not be an afterthought. Emergency controls, independent evaluations and mandatory incident reporting are practical starting points, even if the details still need to be worked out.
The debate between regulation and innovation is unlikely to disappear anytime soon. What is becoming harder to ignore, however, is the need for rules that match the growing influence of artificial intelligence.
Governments have spent years trying to understand what AI might become. They now face the more immediate task of deciding how to manage what it can already do.
