August 30, 2026, (Inside AI) — The gravest dangers from artificial intelligence may not arrive with a mushroom cloud. They may accumulate quietly through thousands of ordinary decisions, each one reasonable on its own, until the point of no return has already passed.
That warning comes from Dr Simon Nieder, a correspondent from Chesterfield, Derbyshire, responding to recent commentary on global AI risk. He argues that the “AI Hiroshima” metaphor misleads because it assumes a single, dramatic moment of catastrophe. Real harm, he says, is more likely to unfold gradually.
“AI may one day behave in ways we cannot control, but many of the gravest harms could happen without any dramatic moment when ‘the AI takes over’,” Dr Simon Nieder, correspondent
His letter, along with others from Dr Anthony Harris and David Kyler, forms part of a widening debate about whether fragmented national efforts can ever match the systemic nature of AI threats.
Quiet Failures Outpace Cinematic Collapse
Nieder points to scenarios where AI assists in designing a pathogen, finding a vulnerability in critical infrastructure, or improving a weapons system while humans still make the final call. The danger is not a rogue machine seizing control. It is the steady erosion of human authority through incremental delegation.
“Other harms may build almost invisibly through thousands of ordinary decisions - more autonomy here, one safeguard removed there, another consequential task handed over because the system has worked well so far,” Dr Simon Nieder, correspondent
By the time the danger becomes obvious, the important decisions may already have been made. That is why he argues international agreement should start with narrow, concrete limits rather than grand philosophical alignment. Countries do not need to agree on superintelligence or extinction probability to agree that some doors should never be opened by AI alone.
Weapons systems, critical infrastructure, and biological synthesis are obvious starting points. Minimum safeguards could include clear human authority for consequential actions, records of who authorised what, and international sharing of serious failures and near-misses.
A Fifty-Year Debate Returns
The current warnings echo a much older conversation. Dr Anthony Harris of Emmanuel College, Cambridge recalls the Serbelloni group, led by Donald Michie at Edinburgh University, which was weighing the social dangers of AI as early as 1972. The technology’s roots stretch further back: Frank Rosenblatt’s perceptron in 1957, the Dartmouth workshop in 1956, and Alan Turing before that.
What is new is the hardware and software infrastructure driving AI forward at terrifying speed. Harris notes that the Lighthill report of 1973 triggered an AI winter in the UK, shifting the field’s centre of gravity across the Atlantic. Donald Michie, who had worked alongside Turing at Bletchley Park, argued forcefully against the cuts in the televised Lighthill debate that followed. The quip attributed to him, perhaps apocryphally, “Let there be Lighthill. And there was darkness,” still resonates.
Harris, a double-hatted academic in computer science and humanities, argues for a return of a “Serbelloni debate” that involves both disciplines. The historical lesson is clear: when one nation retreats from AI research, others advance, and the governance conversation fragments.
David Kyler of Saint Simons Island, Georgia points to an international effort seeking broad political support for a global AI treaty. That effort carries forward an address at the inaugural UN Global Dialogue on AI Governance in Geneva last month, where the secretary general called for human control of AI as its applications spread in weaponry and subversive, prejudicial public surveillance.
Kyler warns that fragmented approaches will cause crucial delays. AI, corporate agendas, national defence goals, and the proliferation of energy-intensive datacentres are entangled in interlocking systemic relationships. Disjointed incrementalism cannot resolve them, especially as the pace of innovation accelerates.
He also flags a troubling diversion: mounting bipartisan opposition to datacentres is drawing political attention away from controlling AI itself. The analogy is treating a serious symptom while ignoring the critical disease causing it. A multifaceted strategy is needed, recognising that the sooner rigorous AI guardrails are established, the fewer datacentre harms will be suffered.
Like climate change, the threats of advanced AI are inherently global. The letters converge on one point: the right battle is not the most cinematic one. It is the quiet, unglamorous work of setting limits before the limits are gone.
