Is regulation enough to stop AI systems killing us all?

In a letter to The Guardian, Martyn Thomas, emeritus professor of IT at Gresham College, questions the efficacy of current regulatory approaches toward frontier AI models. Following OpenAI's recent decision to scrap a model release due to safety concerns, Thomas argues that existing calls for oversight lack a concrete methodology. He draws a parallel to safety-critical engineering sectors like aviation and nuclear power, where developers must provide rigorous 'safety cases' proving that catastrophic accidents are statistically improbable. Thomas contends that AI developers have failed to produce similar evidence or detailed risk analyses, despite the existential risks they acknowledge. He concludes that policy must be grounded in rigorous, fact-based engineering analysis rather than speculative optimism, suggesting that current regulatory frameworks are insufficient to guarantee the safety of advanced AI systems.