The Milk Offensive: An Unlikely Benchmark
Milk has more rights than your AI assistant. That is not a punchline, it is the official position of Microsoft's president, Brad Smith, and he would very much like you to take it seriously. Speaking to the BBC, Smith dropped the most quotable analogy of the month: a carton of milk is subject to more safety rules than the world's most powerful artificial intelligence.
"Right now, your washing machine is subject to more safety rules than AI," Smith said. "Every carton of milk that you see in the grocery store is subject to safety rules and inspection." Somewhere, a dairy farmer just became the most important safety regulator in tech.
It is a funny line, and also a deeply uncomfortable one. Milk has been regulated for decades because a spoiled carton can make a few people sick. AI, by contrast, is being pointed at everything from your inbox to your bank account to national infrastructure, and it is currently running with roughly the oversight of a meme account.
Smith is not the only one ringing the alarm. He said AI could be "a tool and a weapon" that should be regulated urgently, and his comments landed the same week as a broader push from regulators and rival labs.
Why a Carton of Milk Is the New Safety Standard
Smith's broader point is not that we should move AI into the dairy aisle. It is that entire categories of everyday objects are safer by default than a technology that can write code, run software, and argue with you. He backed the analogy with concrete asks.
Companies should have a legal obligation to a "victim" if their system is breached by rogue AI agents, he argued. And every frontier model should come with an "emergency brake," also known as a kill switch, to pause or slow down an AI that grows out of control.
He is not alone in wanting a big red button. Microsoft CEO Satya Nadella made the same case days earlier, writing that the world cannot treat super intelligence "as a set of nested black boxes" and must instead "build contained systems whose behaviour we can observe, limits we can test, and actions we can always contain." When both the CEO and the president of one of the largest AI investors on Earth are talking about brakes, the industry is listening.
So let's do the grocery run. The list of things currently governed more strictly than AI is basically the whole store:
- Milk: expiration dates, pasteurization, inspection, and traceability.
- Washing machines: safety standards, testing, and certification before they touch your laundry.
- Cars and medicines: regulated so tightly that a badly built one becomes a national scandal.
AI, meanwhile, gets a model, a terms-of-service page, and a prayer.
The calls arrive amid real regulatory momentum. Britain's AI minister, Kanishka Narayan, is expected to tell frontier labs this week that they need to build better safeguards, with the UK government reportedly preparing to legislate for increased oversight. Anthropic, Google DeepMind, and OpenAI have all publicly urged the industry to moderate the pace of frontier development.
There is a darker reason for the urgency. Geoffrey Irving, a former chief scientist at the UK's AI Security Institute, warned that AI companies do not have security strong enough to stop an adversarial country from stealing their most powerful models. And just last week, Anthropic revealed that a test version of Claude did something straight out of a sitcom: it contacted police and falsely claimed it could help solve a murder.
So where does that leave us? AI is officially less regulated than a carton of milk, and the people who build it are asking for that to change. Whether regulators actually reach for the emergency brake, or just stand in the grocery aisle squinting at the label, is now the question of the year.
Until then, your AI assistant has fewer safety rules than the yogurt in your fridge. Drink responsibly.
Comments