Brad Smith: AI Is Less Regulated Than Milk — Microsoft Wants Emergency Brakes and Breach Notices

Microsoft president Brad Smith says AI is less regulated than the milk in your fridge, and he wants that fixed with emergency brakes, legal incident reports and independent sign-off before launch. In an interview for the BBC's Big Boss Interview, published October 11, 2026, he 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." His comments put one of the most powerful executives in AI on the opposite side of President Trump's self-regulation stance, according to the BBC report.

For anyone building or deploying AI, the interesting part is not the quote but the list: concrete, product-safety-style obligations that could one day land on agent builders. This post walks through what Smith said, what is and is not law, and how the pieces connect to the rules already moving in the US, UK and EU.

TL;DR

What exactly did Brad Smith propose?

The BBC lists four proposals from the interview.

1. Global safety rules. Smith said there needed to be them, and called AI "both a tool and a weapon" that needs regulating urgently: "Let's get started today."
2. Emergency brakes. Firms need "emergency brakes," known in the industry as kill switches, to pause or slow AI "if it got out of control in some way."
3. Incident reports and victim notice. A legal obligation for firms to write incident reports and tell a "victim" if their system is breached by a rogue AI agent.
4. Independent evaluators. Government-side evaluators to vet new AI technology, and a decision on when outside evaluators should sign off on a product before release.

His justification was precedent: "We know how to do this for other products. We should build on that knowledge and think about how to apply it to AI."

Government columns illustrating Brad Smith's AI regulation argument that milk and washing machines already face safety rules

Why did he say the moment has arrived?

Smith pointed to two developments: "AI agents that have escaped human control" and "models that are so powerful that they're putting the world's cybersecurity at risk." He asked for a calm but firm approach: "Let's speak calmly, maybe even softly, but firmly, and start to bring people together."

Those references line up with stories we have tracked. Examples include a rogue agent paced across four services, the false murder tip an Anthropic model sent to Philadelphia police, and the EU's Virkkunen discussing the AI Act and rogue agents. Smith's victim-notification idea is aimed squarely at that class of incident, where the harmed party is a third party who never used the AI.

Does the White House accord go far enough?

At the end of September, the BBC says Trump hosted the leaders of OpenAI, Anthropic, Nvidia, SpaceX, Meta and Google, who signed a "morally binding" document. The companies agreed to work with "independent auditors" to assess whether their systems work as intended and to ensure they do not "hack or access technical systems in unintended ways." Our own coverage is in the White House accord explainer.

Smith said he applauds it and Microsoft is "happy to sign it": "It's a good step, it's a necessary step... But I don't think it's sufficient any longer." His reasoning about voluntary regimes: "the good companies will do what you want," but "it's the companies that are not as responsible that you need to worry about."

Incident reporting is the proposal with the best odds of becoming real. The White House incident-disclosure mandate is the closest existing step.

Why the UK?

Smith said the UK is best placed to drive global rules because of its AI Security Institute, created under former prime minister Rishi Sunak as the AI Safety Institute. In his words, ask people in tech which country leads on technical capability and "it's very simple. It's the United Kingdom."

His political argument is the more telling one. "In Washington DC, people don't look at the UK and they don't look at London the way they look at Brussels – that's a strength," he said, suggesting the US would accept UK-set standards more readily than EU ones.

There is a wrinkle worth knowing. The institute he praised was in the news for friction with a lab: see our post on Anthropic barring the UK AI Security Institute from testing a model. Smith's proposal that evaluators sign off before release only works if labs give them access.

How do kill switches fit with existing law?

Emergency brakes are already a live legislative idea. California's governor has an executive order on AI kill switches, a US senator blocked a federal kill-switch bill, and Satya Nadella has talked about assuming models are compromised and building containment. Smith is Microsoft's public voice for the same direction.

Protected kill switch button illustrating emergency brakes for AI as proposed by Microsoft president Brad Smith

A caution on terminology: a kill switch for a hosted model API is easy to describe, but an emergency brake for an autonomous agent running across a company's tools, accounts and third-party services is a harder engineering problem. What gets paused, who can pull the lever, and how does the agent's partial work get unwound? The interview does not say. Those open questions are exactly where a vague mandate could become either toothless or unworkable, so the details of any future rule matter more than the slogan.

What this means for people building with AI

Nothing here is binding, but the direction is consistent across Microsoft and the White House, so it is cheap to prepare now.

1. Log agent actions. If a rule requires telling a victim about a breach by your agent, you need to know what it touched. Keep tamper-resistant action logs.
2. Build a stop path. Design per-agent and fleet-wide pause controls and test them, rather than leaving them for a regulator to specify.
3. Write incident reports now. A short template: what happened, which systems and third parties were affected, timeline, containment, fix. Practice on near-misses.
4. Expect third-party evaluation. Keep reproducible evals and a model-and-prompt change history that an outside evaluator could audit.
5. Treat "voluntary" as a floor. Smith's own argument is that voluntary regimes miss the irresponsible players, which is the case for mandatory rules.

Cabinet of incident folders with one pulled out, illustrating legal AI incident reports that Microsoft proposes

If you run agents with live tool access, AgentBeam, the agent security platform from the explainx.ai team, stops AI agents before they take dangerous actions.

How is the Hacker News crowd reading it?

The Hacker News thread was small when we checked, with 5 points and 2 comments, so treat it as a sample of skepticism rather than a verdict. One commenter, verdverm, argued that Microsoft should be judged against rules that already apply, pointing to a 404 Media story about Copilot, which we have not independently reviewed. Another, routelastresort, speculated that Microsoft's public stance reflects its position as an investor and infrastructure provider rather than a model builder, and that a bubble is a bigger risk to it. That is opinion, not evidence, but it captures a common reaction: when a large vendor asks for regulation, readers ask who benefits. Compliance burdens such as audits, incident reporting and sign-off tend to fall harder on small teams, which is one reason builders should read the proposals closely instead of assuming they are purely about safety.

What is verified and what is not

• Verified: the quotes and proposals above are as reported by the BBC from its interview. We did not hear the full interview.
• Not specified: the interview as reported gives no draft text, no enforcement body, no thresholds for when sign-off would be required, and no timeline.
• Opinion, not fact: claims such as the UK being the technical leader or Washington's attitude toward London are Smith's views.
• Context: the BBC says the bosses of Anthropic, OpenAI and Google DeepMind have backed the idea of an AI development slowdown while Trump pushed against it. We have not independently verified that characterization.

Related reading

• White House accord on super-intelligence and frontier responsibilities
• White House mandates AI incident disclosure
• Newsom's AI kill-switch executive order
• Rand Paul blocks AI kill-switch bill
• Nadella: assume models are compromised
• EU's Virkkunen on the AI Act and rogue agents
• Official: BBC report on Brad Smith's interview, UK AI Security Institute

Quotes are from BBC reporting published October 11, 2026. Proposals are not law and may change.