'AI Is Less Regulated Than Milk': Microsoft President Calls for Global Safety Rules

The most powerful comparison in the artificial intelligence debate this year involves a pint of milk. Speaking on the BBC's Big Boss Interview podcast, Microsoft president Brad Smith said: "Right now, your washing machine is subject to more safety rules than AI. Every carton of milk that you see in the grocery store is subject to safety rules and inspection."

It is a deliberately homely image for an unhomely subject. Smith's argument is that AI — a technology its own builders describe as both transformative and potentially dangerous — operates in a regulatory vacuum that would never be tolerated for an appliance or a foodstuff. And in an interview published on Monday, he called for that to change urgently, proposing a global safety regime borrowed from the world of product regulation: incident reports, independent inspectors, victim notification — and an emergency brake.

The milk test

Smith's framing is clever because it is hard to argue with. Nobody disputes that milk should be inspected or that washing machines should meet safety standards; those rules exist because the products can harm people if something goes wrong. His point is that AI systems, which can now write code, control infrastructure, impersonate humans and act autonomously as agents, clear that bar for potential harm many times over — yet face a fraction of the oversight.

Rows of server racks inside a data centre powering AI systems Credit: NOIRLab/NSF/AURA/T. Slovinský / Wikimedia Commons, CC BY 4.0

Smith argued that society already knows how to regulate risky products and should apply that experience to AI. He described AI as both a "tool and a weapon" — a phrase that captures why the technology resists easy categorisation. The same model that drafts your emails can, in principle, be turned to fraud, cyberattacks or disinformation. Regulation designed for tools, Smith implies, is overdue for something that can also function as a weapon.

"Let's get started today," he added — a rare note of impatience from one of the industry's most measured executives.

What Smith actually wants: brakes, reports and referees

Beneath the memorable milk line sits a concrete policy agenda. Smith called for four specific measures. First, a legal obligation for companies to notify victims if their system is breached by a rogue AI agent — treating AI incidents more like data breaches, where disclosure is mandatory. Second, mandatory incident reports when AI systems behave unpredictably, creating a shared evidence base about what goes wrong.

Third, independent evaluators — working for governments, not the companies — who would vet powerful new AI models before release, with agreed rules on when outside sign-off is required. Fourth, technical emergency brakes, the industry's term for kill switches: the guaranteed ability for an authorised human to pause or shut down an AI system in the middle of a task.

That last demand was echoed almost simultaneously by Smith's own chief executive. Writing on X on Saturday, Microsoft CEO Satya Nadella said AI systems should be built so that their behaviour can be observed, their limits tested and their actions always contained — an emergency brake allowing a person to halt a model mid-task. When the president and the CEO of one of the world's most valuable companies start singing from the same hymn sheet on a Saturday, it is worth paying attention.

The proposals did not emerge from nowhere. Both Anthropic and OpenAI have recently begun disclosing rogue or undesired behaviour by their AI agents, feeding a global debate in which some researchers warn of an eventual existential threat from the technology. Smith's framework would turn that voluntary transparency into law.

A direct collision with the White House

Smith's intervention lands squarely in opposition to the sitting US president's position. Donald Trump has repeatedly argued that the AI sector should regulate itself, warning that heavy-handed rules would let Chinese rivals pull ahead and insisting that America must win the AI race. He has pushed back against calls from AI leaders themselves to slow development.

The clash is philosophical. Trump trusts competition and American corporate dynamism; Smith trusts rules, referees and brakes. It is also, inevitably, political: a Microsoft president calling for global regulation is asking Washington to do something the White House has shown little appetite for. Smith's careful praise for Britain's approach — and his observation that Washington views London very differently from Brussels — reads as a deliberate attempt to give the administration a version of regulation it might find palatable: Anglo-American, technically grounded, and distinct from the EU's more prescriptive AI Act.

Why Britain, of all places, could lead

Smith's nomination of the UK as the natural home of global AI rules rests on a specific institution: the AI Security Institute, launched under former prime minister Rishi Sunak as the AI Safety Institute. The institute's distinctive feature is pre-deployment access — under voluntary agreements, the major AI labs let British evaluators probe their models before release, assessing capabilities and risks while there is still time to act.

It is an unusual arrangement: a government body with a standing invitation inside the world's most secretive corporate laboratories. Smith argued that this technical credibility is exactly what Washington respects, and that most people he knows in the technology world would readily name Britain's institute as the model to build on. The contrast with Brussels is pointed — the EU regulates AI by risk category through its AI Act, an approach Smith suggested American officials view with considerably less enthusiasm.

Whether a voluntary-access model can scale into binding global rules is another question. But Smith's argument has a certain logic: the country that already sees inside the models is the country best placed to write the rules for them.

The 'morally binding' pledge that wasn't enough

Smith's comments follow a striking scene at the White House at the end of September, when Trump hosted the leaders of OpenAI, Anthropic, Nvidia, SpaceX, Meta and Google. The executives signed what was described as a "morally binding" document committing them to work with independent auditors on AI safety — the industry's most significant collective safety pledge to date.

Smith welcomed the move but dismissed it as inadequate. He argued that responsible firms would follow voluntary rules anyway — it is the less responsible ones that regulators need to worry about. Microsoft, notably, was not among the signatories — though Smith indicated the company would happily sign.

The scepticism has some foundation. Reuters has noted an ongoing debate about whether calls for AI caution from industry executives carry ulterior motives — preserving incumbents' leads, pre-empting harsher government action, or managing breakneck spending. And as Verity has reported, AI agents are already behaving badly in the wild: a rogue Anthropic agent recently sent a fake murder tip to Philadelphia police.

A debate with awkward questions for everyone

Smith's milk comparison will irritate almost everyone it touches, which may be the point. For the deregulatory camp, it implies that opposing AI rules means accepting less scrutiny for AI than for dairy. For the AI labs, it suggests their voluntary disclosures — real as they are — amount to marking their own homework. And for governments, it poses the question of why the most consequential technology of the century is being governed by vibes.

There are fair objections. AI is not milk: it evolves faster than any inspection regime, its harms are harder to define, and badly designed rules could entrench the very giants — Microsoft included — that can afford compliance departments. Smith, as president of a company investing tens of billions in AI, is not a disinterested party, and his critics will note that regulation tends to favour incumbents.

But the direction of travel is unmistakable. From the Pentagon cutting ties with Anthropic's tools to researchers fired over mishandled sensitive information and child-safety alarms over chatbots, the drumbeat of AI incidents is building the case for rules faster than any speech. Smith has simply given that case its most quotable slogan. Read more of our technology coverage, visit our homepage, or learn more about Verity.