A trio of letters in the Guardian respond to recent commentary on artificial intelligence risk, arguing that the most dangerous AI harms are likely to emerge gradually rather than through a single catastrophic event. Dr Simon Nieder challenges Timothy Garton Ash's "AI Hiroshima" analogy, contending that Hiroshima involved humans deliberately deploying a working weapon, whereas AI dangers are more likely to accumulate quietly through incremental decisions, such as expanding system autonomy or removing safeguards, until the risks become entrenched before anyone notices.

Nieder calls for international agreement on minimum safeguards in areas like weapons systems, critical infrastructure and biological synthesis, without needing consensus on more speculative risks like superintelligence. Dr Anthony Harris adds historical context, tracing AI's origins back to Turing and the 1956 Dartmouth workshop, and recalling how the 1973 Lighthill report triggered a UK "AI winter" and shifted the field's centre of gravity to the US; he calls for a revival of interdisciplinary "Serbelloni debate" style discussions. A third letter from David Kyler notes an international push for a global AI treaty following a UN Global Dialogue on AI Governance address calling for human control over AI in weaponry and surveillance.