AI Faces Less Oversight Than Milk, Says Microsoft President Brad Smith

Milk is pasteurised, labelled, inspected and traced to its source. Washing machines must meet electrical safety standards before they can be sold. Artificial intelligence, the most transformative technology of the decade, faces none of that. That is the striking comparison Microsoft president Brad Smith made this week, arguing that one of humanity's most powerful inventions operates under lighter rules than the contents of your fridge.

Speaking to the BBC's Big Boss Interview, published on 11 October, Smith called for global safety rules with real enforcement: emergency brakes to halt runaway systems, a legal obligation to notify victims when AI goes rogue, and independent evaluators who sign off on new technology before release. His remarks put one of the most influential figures in the AI industry at odds with the Trump administration's preference for self-regulation — and opened a new front in the battle over who writes the rules for the machines.

The milk comparison

Smith's central claim was disarmingly simple. "Right now, your washing machine is subject to more safety rules than AI," he told the BBC. "Every carton of milk that you see in the grocery store is subject to safety rules and inspection." The comparison is deliberate: ordinary products pass through layers of testing, certification and liability before reaching consumers, while AI systems can be trained on billions of data points and deployed to hundreds of millions of users with minimal formal oversight.

He described AI as both a "tool and a weapon" that urgently needs governing, urging policymakers: "Let's get started today." The framing echoes a long-running complaint from safety researchers — that the technology is being scaled faster than the rules that govern it — but it is unusual to hear it stated so bluntly by a sitting executive of one of the companies doing the scaling.

The demands for tighter control have been growing. Leaders of the US-based AI giants, including Anthropic, OpenAI and Google DeepMind, have all backed the idea of slowing down AI development at various points, amid warnings from experts that the technology could pose a threat to humanity. Smith's intervention adds Microsoft's institutional weight to that camp.

Three proposals with teeth

What made the interview notable was not just the rhetoric but the specificity. Smith laid out three concrete mechanisms borrowed from product safety law. First, "emergency brakes" — known in the industry as kill switches — that would let operators pause or slow AI systems if they begin to behave unpredictably or dangerously.

Second, mandatory incident reporting: a legal obligation for firms to write up formal "incident reports" and notify victims if their systems are breached by a rogue AI agent. The idea treats AI failures like aviation accidents or data breaches — events that must be documented and disclosed rather than quietly patched.

Third, independent evaluation before launch: outside, government-backed evaluators who would assess new AI technology and sign off on it before release. "We know how to do this for other products," Smith said. "We should build on that knowledge and think about how to apply it to AI."

Together, the three proposals would move AI governance from voluntary commitments to enforceable obligations — the crucial distinction Smith kept returning to. Promises are fine, his argument runs; liability is what changes behaviour.

A clash with the White House

Smith's call for binding global rules puts him in direct tension with President Donald Trump, who has argued that the sector should regulate itself so that American firms do not lose out to Chinese rivals. The administration's approach has favoured voluntary industry pacts — such as the White House Accord on Super Intelligence signed by leading tech chiefs last month — over new statutory restrictions.

Smith did not reject the accord outright. According to reporting on his remarks, he said he was "happy to sign it" but considered it "not sufficient any longer" — a polite way of saying that self-policing cannot restrain the industry's less scrupulous actors. The concern is the familiar one with voluntary regimes: responsible companies sign, irresponsible ones do not, and the floor keeps falling.

The disagreement reflects a genuine strategic split. Washington fears that heavy regulation will slow American labs while Chinese competitors sprint ahead. Smith's camp fears that no regulation invites the kind of accident — or abuse — that destroys public trust and invites a far harsher crackdown later. Both sides claim to be protecting American leadership; they differ on what threatens it.

Where the rules stand today

Smith is right that the regulatory landscape is patchy. In the European Union, the AI Act — the world's most comprehensive AI statute — is being phased in, with high-risk systems facing conformity assessments, transparency duties and human oversight requirements. In the United States, there is still no comprehensive federal AI law; governance is a patchwork of agency guidance, executive orders and state-level experiments.

Britain has taken a third path: light-touch, principles-based, with sector regulators applying existing law. Smith singled out the UK as best placed to drive global standards, pointing to its AI Security Institute — launched under former prime minister Rishi Sunak's government as the AI Safety Institute — as the natural convening body for international work. The endorsement is a notable vote of confidence in Britain's institutional bet that it can lead on AI governance without legislating as heavily as Brussels.

The gap Smith describes is real enough: a dairy farmer faces more routine inspection than a lab training a frontier model. Whether his proposed remedies are proportionate — and who would staff an army of independent evaluators — remains the open question.

A messenger with a stake

Critics will note the source. Microsoft is one of the largest deployers of AI on the planet, with its Copilot assistant woven through Office products and its cloud infrastructure rented to countless model builders. A call for regulation from a company that profits from the technology invites the charge of regulatory capture — incumbents often favour rules that entrench their advantage while raising barriers for challengers.

There is also the question of enforceability. Emergency brakes sound reassuring, but defining the threshold at which a system must be halted — and who gets to pull the lever — is fiendishly hard for systems whose failures emerge unpredictably. Incident reporting duties work only if companies can detect rogue behaviour in the first place. And independent pre-release evaluation of every significant model would require a regulatory apparatus that does not yet exist anywhere in the world.

Smith's answer is that imperfection is no excuse for inaction. The product-safety regimes he invokes — food inspection, electrical standards, aviation reporting — were all built after accidents, not before them. His argument is that AI should be the exception: regulated in anticipation of harm rather than in its aftermath.

Why it matters

The interview lands at a pivotal moment. Governments on three continents are deciding whether AI's next chapter will be written by statute or by corporate pledge. Smith, speaking from inside the industry rather than outside it, has effectively told regulators that the current bargain — voluntary commitments plus presidential summits — is not enough.

Whether his proposals become law is uncertain. But the milk comparison has already done its work: it has given the regulation debate a vivid, domestic image of the gap between what we inspect and what we do not. The question now is whether lawmakers — in London, Brussels or Washington — are willing to treat a chatbot with the seriousness they already give a carton of milk.