Andrew Bailey: Why AI regulation should wait for better testing

The 'No Regulation' Stance

Listen, everyone from Silicon Valley to the White House is obsessed with AI right now, but Andrew Bailey—Governor of the Bank of England—is pumping the brakes on the government's approach. In his first-ever Substack piece, he dropped a hot take: trying to regulate AI immediately is a major L. He argues that we shouldn’t be rushing into a rulebook just yet.

Stress-Testing > Rules

Bailey isn't saying we should just let AI run wild with no oversight. He’s all about "rigorous" testing. His vibe is that we need to find where these models break—because, real talk, they will fail—before we try to legislate them. He’s pushing for a framework where we can identify vulnerabilities and set guardrails to keep risks from spiraling.

"That is not evidence that testing has failed, rather it is evidence of why testing is necessary," Bailey noted. Basically, expecting AI to behave perfectly right out of the gate is a setup for failure.

The Big Picture

This comes after a week where the AI conversation has been hitting a fever pitch. On one side, companies like OpenAI and Anthropic are lowkey scared, suggesting we need to slow down development. On the other side, you’ve got President Trump, who is highkey leaning into the AI arms race against China. Trump recently got these tech CEOs to sign a "morally binding" deal to keep their own tech safe, but Bailey seems to think we need more than just vibes and promises.

In the UK, the AI Security Institute is already working on these standards, but Bailey says the pace needs to pick up. He isn't ruling out regulation eventually—it’s just not the right starting point while the tech is still this unpredictable.

Why it matters

If the head of the Bank of England thinks we aren't ready to regulate, it’s a sign that the industry is still in a "wild west" phase. For investors and workers, this means the rules of the game are going to stay fluid for a while. Don't expect a sudden government crackdown; expect a long, messy period of trial and error as these systems get stress-tested.