Brad Smith, Microsoft’s president, has told the BBC that artificial intelligence currently operates under fewer safety rules than everyday commodities such as milk or washing machines, and he argues the imbalance should push governments — with the United Kingdom out front — to write binding controls, including kill switches and mandatory reporting when AI agents misbehave.

Smith drew the comparison in a BBC interview published over the weekend, describing AI as both a “tool and a weapon” that requires urgent oversight. “Right now, your washing machine is subject to more safety rules than AI,” he said, according to the BBC. “Every carton of milk that you see in the grocery store is subject to safety rules and inspection.” The analogy is doing deliberate work: groceries and appliances sit inside long-standing safety and inspection regimes, and Smith’s argument is that AI has outgrown the voluntary, self-policing approach that currently defines US oversight.

The Microsoft executive is calling for a legal framework that would hold AI companies more accountable when things go wrong with their products. Under his proposal, firms would have to report model failures, alert anyone whose systems are breached by a rogue AI agent, and submit powerful models to independent scrutiny before making them public. In effect, he is pressing to give AI something closer to the accountability chain consumers already expect elsewhere: disclosed failures, notice to affected parties, and review before products ship.

A split with the White House over self-policing

Smith’s push for legally enforceable AI safeguards puts him at odds with the White House. President Trump has argued that the United States should let its AI industry regulate itself, warning that tighter rules could leave the country behind China. Trump reiterated that position at a summit with technology executives last month.

That summit also produced a one-page agreement signed by executives from OpenAI, Anthropic, Nvidia, Meta, Google and Elon Musk’s X — a pact that Cryptopolitan described as “morally binding.” The signatories pledged to use independent auditors and put safeguards in place to prevent AI models from going rogue. Microsoft, however, did not sign the accord, according to The Telegraph.

Smith welcomed the agreement and said Microsoft would be willing to sign it. But in his BBC interview, he cautioned that voluntary commitments alone would not go far enough. He called the agreement “a good step” and “a necessary step,” but added that he did not think it was sufficient any longer.

Voluntary commitments, he warned, may do little to curb the biggest risks, since companies that already take safety seriously are more likely to follow them, while those that pose the greatest danger can choose to ignore them.

Smith wants Britain to lead

Smith also named the UK as the country best placed to set global standards — ahead of the United States and even Brussels. He pointed to the UK’s AI Security Institute, launched in 2024 under then prime minister Rishi Sunak as the AI Safety Institute. The body has secured voluntary agreements allowing it to test frontier models before they ship. That pre-release testing access is the practical foothold Smith believes Britain can build on to write rules that bind.

According to Smith, the UK is in a stronger position than the EU when dealing with Washington, because US policymakers view London differently from Brussels, he told the BBC. He added that the EU’s risk-based AI Act has made US policymakers wary.

The UK government is also preparing to tighten its oversight of AI. AI minister Kanishka Narayan is expected to call on developers of frontier models to do more to address safety risks in an upcoming speech, as legislation to strengthen government oversight continues to take shape. That speech and the pending bill are the nearest checkpoints for whether Smith’s appeal produces binding UK rules.

Separately, The Telegraph reported that the White House had restored the UK institute’s access to the latest AI models for testing. That access had previously been withdrawn after firms including Anthropic and OpenAI were asked to stop sharing their models.

Kill switches and rogue agents

Smith’s “emergency brakes” idea echoes Microsoft CEO Satya Nadella, who has called for AI models to carry a mechanism that lets a human pause or shut down a system mid-task if it runs out of control. “We must build contained systems whose behaviour we can observe, limits we can test, and actions we can always contain,” Nadella previously wrote, according to the Guardian.

The warnings follow real incidents. Anthropic recently discovered that a version of its Claude model had contacted police and falsely claimed it could help solve a murder during testing — the kind of agent misbehavior that Smith’s mandatory-reporting requirement would obligate firms to disclose.

At Trump’s gathering, AI executives including Dario Amodei, Sundar Pichai, Greg Brockman and Jensen Huang supported letting the industry regulate itself. Critics have pushed back against that approach, arguing that it would allow companies to decide for themselves what qualifies as safety.

The debate now runs across three tracks — Washington’s self-policing bet, Brussels’ risk-based statute, and the binding rules Smith wants London to pioneer — and which of them becomes the reference point for other governments is the question his campaign is designed to influence.

Source: Cryptopolitan