Regulating AI ‘not the right place to start’ says Bailey
Andrew Bailey, Governor of the Bank of England, has argued that regulating artificial intelligence should not be the immediate priority, calling instead for "rigorous" testing to identify vulnerabilities and establish safeguards. Writing on Substack, he acknowledged that AI risks are "real and increasingly significant" but emphasised that development should not be prohibited, stating instead that there must be systems for intervention and boundaries within which AI operates.
This position reflects an ongoing debate about managing AI's rapid development. Whilst AI firm executives including those from Anthropic and OpenAI have recently called for slowed development and internationally coordinated safeguards—with OpenAI declining to release its latest model citing safety concerns—other voices, including President Donald Trump, have rejected a slowdown. The UK has established an AI Security Institute to conduct testing, and Bailey emphasised that this testing work should accelerate "with a degree of humility", acknowledging that models will sometimes behave unexpectedly. He stressed that testing should not be viewed as an alternative to future regulation, but rather that regulation is not the right starting point.
- Testing and safeguards should precede formal AI regulation, says Bank of England Governor.
- AI risks are real but development shouldn't halt; testing must accelerate with humility.
- UK's AI Security Institute leads testing efforts to establish standards and safety boundaries.
Both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
A testing-first approach allows policymakers to ground regulation in empirical evidence of actual AI risks rather than theoretical concerns, preventing premature rules that may stifle beneficial development or become obsolete as the technology evolves. Continued development with appropriate oversight and safeguards is necessary to identify safety problems and create applications that address societal challenges; regulation that arrives too early may entrench outdated approaches or impose requirements that prove misguided once the technology is better understood.
The case against
Waiting for comprehensive understanding before establishing regulatory frameworks inverts the precautionary principle and risks embedding powerful AI systems in critical infrastructure before safeguards are in place. Precedent from pharmaceuticals to aviation shows that proactive regulation during development, not retrospective regulation after deployment, protects public interests. Regulatory frameworks and rigorous testing are not mutually exclusive; they can proceed in parallel as in other industries, but without binding rules and accountability mechanisms in place now, voluntary safeguards prove insufficient to constrain the concentrated power of a few companies developing increasingly capable systems.