Expert comment: growing calls for AI regulation

Professor Richard Whittle, Professor of AI and Public Policy at the University of Salford, comments: “The AI and existential risk debate is back on the agenda. Dario Amodei has called for the industry to slow the pace at which it improves model capability, warning that a swarm of agents could do enormous damage within the year. Jacob Coxon, a departing Anthropic researcher put the odds of extinction this decade at one in ten. This is terrifying and governments are looking to regulation and slowdowns. The 2026 International AI Safety Report, written with more than a hundred independent experts, is more careful. It finds early signs of the relevant capabilities but not at levels that would permit a loss of control.

“These warnings should not be ignored, they should specifically not be ignored by those calling for AI adoption at any cost. The existential risks of future AI are clear and we should be debating them, but the more mundane risks, the economic ones are present in our economy now and are being ignored for the sake of unfettered AI readiness and a desperation for growth and innovation.

“Alignment failure is a frightening possibility we need to navigate and protect against, but for an economy deciding what to do with AI, the issue is not superintelligence. It is AI pricing, labour effects and the looming burst of the AI bubble, these are the here and now and we need to factor them into our planning.

“Five hyperscalers are expected to spend upwards of seven hundred billion dollars on capacity this year, financed increasingly by debt and by contract-backed borrowing. The prices at which councils, universities and businesses are adopting these tools are designed to gain market share rather than the cost of the service. When this market adjusts, whether through a correction or simply through investors asking what they are getting, those prices increase. 

“An organisation that has rebuilt a process around a tool it does not own is out of luck. It needs to pay whatever is asked of it. Doubly so if, in the cheap AI years, it has neglected talent pipelines, prioritising borrowed short term cost savings over long term sustainability.

“We should, of course be asking about capabilities, risk and safety. Experts in AI alignment are worried. But we also need to ask what terms of adoption we are prepared to accept. Adoption regardless of the economic risks, readiness over resilience? We need immediate thinking about the looming AI risks and how to mitigate them.”

Craig Smith, Lecturer in Law, University of Salford, says: “The latest warnings from leading AI developers should be taken seriously, but not uncritically. Claims that AI could escape human control, compromise the internet or threaten humanity are extraordinary. They require evidence proportionate to their significance, alongside scrutiny of the corporate interests surrounding them.

“Professor Andy Miah at the University of Salford identifies an important contrast with climate change: “With climate change, it was the scientists, with vast amounts of evidence. With this, it is former AI employees with their anecdotes from the inside.” AI development remains unusually opaque, akin to a black box. But insider testimony or a whistleblower type scenario cannot substitute for independent research, transparent testing and rigorous data about how widely these concerns are shared.

 “The current news also exposes a central paradox in AI governance. The companies creating the risks possess much of the technical knowledge needed to understand them, yet they also have commercial interests in shaping how those risks are described and regulated. The sharp market reaction demonstrates the economic power carried by statements from a small group of executives.

“Nor is the call for action new. The Bletchley Park AI Safety Summit in 2023, the AI Seoul Summit in 2024 and the Paris AI Action Summit in 2025 all recognised that AI risks cross national borders. They produced declarations and voluntary corporate commitments but limited enforceable accountability. The Council of Europe’s Framework Convention represents an important move towards binding international law, although it is not a comprehensive global regime and depends upon national implementation. While the EU itself has an EU AI Act, this is not an international agreed framework. 

“The response should therefore go beyond asking companies to slow down voluntarily. Governments need independent access to frontier models, mandatory safety evaluations and incident reporting, enforceable human-rights protections, and clear corporate accountability when harm occurs. Yet all this slows down the development of AI and world leaders in the AI race have already commented on their aims to win said race. 

“We should neither dismiss these warnings as science fiction nor accept them as settled science. The legal priority is to ensure that uncertainty does not become an excuse for inaction, or for handing governance to industry. A global technology cannot be governed safely through fragmented national rules and voluntary promises written by the corporations developing it.”