Regulating AI 'not the right place to start' says Bailey

The Bank of England’s governor has taken a clear line on AI: don’t start with regulation. In his first article for Substack, Andrew Bailey argued that the first priority should be “rigorous” testing to find weak spots and build safeguards that can contain the damage if systems go wrong.

Bailey did not downplay the risk. He called the dangers around AI “real and increasingly significant,” but said the technology should not be stopped or banned. His view is that the benefits are “immense,” and the real task is to set boundaries and create a way to step in when systems cross them.

That puts him in the middle of a debate that has sharpened in recent weeks. Leaders at firms including Anthropic and OpenAI have pushed for development to slow down and for a co-ordinated international effort on risks and safeguards. OpenAI has already said it will not release its latest model because of safety concerns.

Donald Trump has rejected the idea of a slowdown, saying the US is ahead of China in the AI race and wants to stay there. But after a summit with AI firms including OpenAI and Anthropic, he said executives had signed what he called a “morally binding” document that would act as a form of protection against AI risk, with companies themselves responsible for safety.

Bailey said the UK is already doing “important work” through the AI Security Institute, which he sees as part of a broader set of standards. Still, he wants the pace to pick up. Testing, he said, will produce failures and models will behave unexpectedly. That is not proof the testing has failed, just proof that it is necessary. Regulation may come later, he said, but “is not, in my view, the right place to start.”