Bank of England governor rejects rushing legislation, saying technical vulnerability testing must precede AI regulation

Bank of England Governor Andrew Bailey believes that starting with legislation to regulate artificial intelligence is not the proper path for dealing with its growing risks, calling instead for giving top priority to rigorous testing to uncover vulnerabilities and building technical safeguards that contain potential threats. Bailey, in his first article published on the Substack platform, said that the risks surrounding the technology are real and are continuously increasing, but he also emphasized the need not to halt or ban development in light of the huge returns expected, while calling for the establishment of a clear system for technical intervention and the drawing of precise boundaries within which models operate.

Technical tests reveal unexpected pathways and sometimes fail, yet that failure in itself is conclusive proof of the necessity of testing.The central bank governor builds on this view to state that models stumbling during pilot testing does not indicate a failure of regulatory standards, but rather demonstrates the need to accelerate field experiments while maintaining a degree of scientific humility before the complexity of intelligent systems. Bailey explains that the UK AI Safety Institute is carrying out important work to establish these standards, stressing that testing is not a final substitute for legal frameworks that may develop over time, but that ready-made laws at this stage are not the correct starting point.

The most prominent warning Bailey raises focuses on what are called advanced interface models, the highly autonomous learning systems. This category in particular carries a real risk of turning into a closed loop in which the model governs itself gradually unless decisive external mechanisms for human and technical intervention are available. This stance comes at a time when international divisions over technology governance are intensifying, as CEOs of companies such as Anthropic and OpenAI call for slowing development and international coordination to assess risks, while the US administration under Donald Trump rejects any idea of slowing down in order to maintain its lead over China, relying instead on a non-binding document signed by major tech firms to assess the safety of their tools internally.

This shift toward prioritising technical testing over legal texts reshapes the priorities of financial and banking institutions in the Gulf, Egypt and the Levant. Central banks and financial supervisory authorities in the region, which usually monitor guidance from the Bank of England and international financial bodies, see in this proposal a direct incentive to invest in experimental testing environments and develop technical testing standards instead of pre-empting the market with rigid legislation that could hinder the adoption of automation and credit solutions. At the level of teams and developers within companies and the regional banking sector, the weight of required competence changes: the operational priority will not be drafting paper compliance clauses, but building teams capable of conducting rigorous penetration tests, measuring the behaviour of intelligent models, and installing stop-points and human intervention that prevent operational processes from slipping into uncontrolled closed loops.

Building strict intervention and control tools precedes the drafting of legal regulations in sensitive operating environments.The technical safety pathway starts with subjecting every experimental model to continuous stress tests within isolated environments before integrating it into decision-making chains, because asset protection and transaction stability are not achieved by intentions or delayed regulatory texts, but by software barriers that have the authority to isolate and intervene at critical moments.