Why Brad Smith Wants London Writing AI’s Safety Rules

Microsoft president Brad Smith says the world needs binding AI safety rules now, and he thinks the U.K. should write them. TechCrunch AI reports that Smith made the case on a recent BBC podcast. He called AI both a “weapon and a tool” that’s putting global cybersecurity at risk, and he said the “moment has arrived” for global safety rules.

He also gave a pointed comparison. Milk and washing machines face more oversight than the companies building frontier AI.

🇬🇧 Why London and not Washington or Brussels

Smith’s argument rests on politics as much as policy. Britain’s AI Security Institute (AISI) launched in 2023, and it already gets access to major U.S. frontier models before they’re released. That pre-deployment testing has quickly made it a leader in AI safety evaluation.

The EU has rules, but Washington sees Brussels as an adversary. The U.S. has the labs but no appetite for regulation. The U.K. sits in between. “In Washington, D.C., people don’t look at the U.K. and they don’t look at London the way they look at Brussels,” Smith told the BBC. “That’s a strength.”

That position is shakier than it sounds. According to TechCrunch AI, President Trump has warned against “globalist” organizations interfering with American companies. Last month he reportedly asked Anthropic and OpenAI to hold new models back from British testers. AISI confirmed on Monday that it can still access and test the latest models, but the pressure is real.

⚖️ Voluntary pledges vs. real rules

The timing matters. Last month Trump hosted leaders from OpenAI, Anthropic, Nvidia, AMD, SpaceX, Meta, and Google at the White House. They signed a “morally binding” document promising to work with “independent auditors” on model safety and capabilities. Dario Amodei and Sam Altman have both called for slowing development after several incidents of AI agents going rogue.

Smith called the pledge a “good step” but said it falls short. His reasoning is simple: voluntary rules only bind the “good companies.” “It’s the companies that are not as responsible that you need to worry about,” he said.

Here’s what he wants instead:

• Breach disclosure: a legal duty to report when rogue AI agents compromise a system
• Pre-release approval: outside evaluators sign off before an AI product ships
• Kill switches: ways to pause or slow AI if humans lose control

🔍 Reading Smith’s track record

Smith isn’t new to this argument. He called for facial recognition regulation back in 2018, and his 2019 book was literally titled Tools and Weapons. He’s been consistent for years.

You should still read his comments in context. Microsoft is one of the best-resourced AI players on the planet, and it can absorb compliance costs far more easily than smaller rivals. Rules that burden “less responsible” companies also raise the barrier to entry. Both things can be true: his proposals make sense, and they also suit Microsoft.

🔮 Where this goes in the next 1-3 years

What stands out is how far the debate has moved. The question isn’t whether frontier AI gets outside testing anymore. It’s who does the testing and whether anyone can enforce the results.

A few likely developments:

• Pre-deployment testing becomes the norm. Even the White House pledge accepts independent auditors. Turning that into law is a much smaller step than it used to be.
• Incident reporting arrives first. Breach disclosure rules already exist for data and cybersecurity, so extending them to AI agent failures is the easiest win politically.
• AISI’s access turns into a bargaining chip. If U.S.-U.K. tensions rise, British testers could lose early model access. That would weaken Smith’s whole thesis.

🛠️ What practitioners should do now

If you’re deploying AI agents, don’t wait for the law to catch up:

1. Log agent actions so you can reconstruct what happened if something goes wrong.
2. Build pause controls into agentic workflows. A kill switch is far cheaper to design in now than to retrofit later.
3. Draft an incident response plan covering AI-caused breaches, not just traditional ones.
4. Watch the vendor commitments. The auditor pledges your model providers signed will shape which evaluations and reports you can actually get from them.

Smith ended on a measured note: “Let’s speak calmly, maybe even softly, but firmly, and start to bring people together.” Whether the U.K. can play that role depends less on London than on how much patience Washington has. You’ll find the full story at TechCrunch AI.