The Governor of the Bank of England, Andrew Bailey, said, “Intensive AI testing should come before official regulation,” adding that authorities should retain room to intervene as risks grow.
Key content
Bailey pointed out that while AI risks are increasing, vulnerabilities must be identified first before drawing up a regulatory framework.
He emphasized that, along with rigorous testing throughout the entire process before and after deployment, clear safety measures and reliable points for human intervention must be put in place.
His remarks came amid a debate between major AI companies and the U.S. government over self-regulation, how to moderate the pace of development, and the direction of future regulation.
Bailey: “AI should be put on the test bench first”
In a BBC op-ed, Bailey assessed that AI risks are “realistic and becoming increasingly serious,” but argued that before deciding how regulatory regimes should work, one must first identify where and how systems can fail. He said, “Starting with regulation is not right.”
Bailey said society must maintain the capacity to set boundaries for advanced AI, and that through thorough testing before and after deployment, vulnerabilities should be exposed and it should be confirmed whether safeguards can actually control accidents. He explained that AI development should not be banned because “the potential benefits are enormous.”
The UK government has already established an “AI Security Institute” dedicated to AI security. While Bailey said he believes the institute contributes to building a scientific and technical foundation for ensuring AI reliability, he added that “the pace of progress needs to be faster.” He also noted that unexpected model behavior is less evidence of test failure than “an example showing why testing is essential.”
Related article: Ripple partners with Brazil’s CSD BR to mirror a fund-holding structure for a public blockchain
The safety debate surrounding OpenAI
Major AI firms such as OpenAI and Anthropic are calling for stronger safeguards and a risk-assessment framework under international coordination as model performance rises rapidly. OpenAI has recently said it will not release its newest AI models due to safety concerns.
Meanwhile, U.S. President Donald Trump rejected calls to slow down the pace of AI development in competition with China, saying, “Whoever wins AI, wins.” After meeting with CEOs of major technology companies at the White House on September 29, he described an AI-related document signed by attendees as a “morally binding agreement” and a kind of “safety mechanism.”
Under this agreement, the participating companies adopt a structure in which they take responsibility for the safety of their own AI technologies. Bailey’s approach differs from the U.S.-style “corporate self-accountability” model in that, before devising a more formal regulatory framework around advanced AI, it must first clarify through testing where public intervention is needed.
In recent weeks, as AI developers, governments around the world, and financial regulators have faced the issue of faster model upgrading and the resulting supervisory gaps, related discussions have accelerated. Bailey’s remarks suggest that, amid the push for international coordination mechanisms demanded by the industry, OpenAI’s decision to keep its models private, and the White House’s stance on company-led self-regulation, the global debate over how to manage AI risk is entering a new phase.
Further reading: Altcoins account for 41% of open interest… spot trading volume is 4 times that of Bitcoin