The takeaway: On the same day 29 countries signed a new AI cooperation organisation into existence in Shanghai, more than 200 Nobel laureates, academics and innovators in Rome signed a declaration urging the opposite instinct: a "coordinated slowdown of frontier AI development," a ban on unmonitored recursive self-improvement, and a treaty keeping AI out of nuclear launch decisions. Two governance visions were put on paper within hours of each other — and they point in opposite directions.

Where it came from

The declaration — titled "Humanity at the Threshold: A Declaration on Artificial Intelligence and Nuclear Weapons" — is the product of the Global Nobel Laureates Assembly on Nuclear Weapons and AI, convened 14–16 July at Borgo Laudato Si', the Vatican's Castel Gandolfo estate. It was signed on Thursday at the Giulio Cesare Hall of the Palazzo Senatorio — Rome's city hall on the Capitoline Hill — by over 200 participants, according to Catholic World Report, which says the event drew Cardinal Baldassare Reina, Rome's mayor Roberto Gualtieri, physics Nobel laureate David Gross and actress Sharon Stone, and describes the text as inspired by Pope Leo XIV's encyclical Magnifica humanitas.

Among the signatories publishing it: Nobel Peace laureates Juan Manuel Santos — the former Colombian president — and Denis Mukwege, whose organisation The Elders carries the full text.

What it actually says

The framing is deliberate and historical: "Humanity stands at a defining moment in its history. More than eighty years after the dawn of the nuclear age, and at the threshold of the age of artificial intelligence, we are presented with an unprecedented challenge." On the nuclear precedent, the document is blunt: the world had a chance to avoid a permanent dynamic of fear — "We failed."

Four principles follow. The most quotable is the third — "An automated system should never make the final decision to launch a nuclear weapon" — backed by a call for "an international treaty prohibiting the reckless integration of artificial intelligence into nuclear command, control, and launch systems so that meaningful human control is maintained."

But the second principle is the one frontier labs will read twice. Beyond asking developers to "transparently publish the principles that guide the behavior of their models" and accept liability for adherence to them, it draws a hard line on self-improving systems: "no organisation should initiate, and no government should permit, fully-automated recursive self-improvement in artificial intelligence systems without the means to monitor, and if needed, to halt such systems."

And the fourth principle asks governments, corporations and international organisations "to enable coordinated slowdown of frontier AI development by establishing shared mechanisms, such as verification and robust internal and external evaluations" — alongside support for the UN's Independent International Scientific Panel on AI, whose report went in front of governments in Geneva earlier this month.

The first principle names the game directly: AI is "an arena of strategic competition, like nuclear weapons before it," and the signatories want "restraint in its most destabilizing applications" — "We must disarm the next arms race, both AI and nuclear, before they define the next century as well."

The same-day counterpoint

It is hard to imagine a sharper contrast with what happened 9,000 kilometres east. In Shanghai, Thursday brought the founding of WAICO by 29 countries, and Xi Jinping's keynote this morning made the pitch explicit: faster, wider, more institutionalised AI development, with concrete pledges to the Global South. The Rome signatories want verification mechanisms and a slowdown; the Shanghai communiqué wants application centres and adoption. Both invoke the UN. Both claim to speak for humanity's common interest.

The honest observation is that neither document binds anyone. WAICO is an institution with a headquarters and no published membership list; the Rome Declaration is a moral statement with 200 distinguished signatures and no enforcement mechanism. But moral statements have a track record in the nuclear domain the declaration deliberately invokes — the Pugwash conferences and the Russell–Einstein manifesto shaped treaty politics for decades. Whether "Humanity at the Threshold" becomes that kind of reference point, or a well-attended afternoon on the Capitoline, depends on what governments do with its treaty call.

What we're watching

Whether any government formally picks up the nuclear command-and-control treaty proposal; whether the UN Scientific Panel cites the declaration; and whether any frontier lab responds to the recursive self-improvement line — a boundary the Global Call for AI Red Lines (whose signatories include 15 Nobel Prize and Turing Award recipients) sketched in 2025 as a prohibition on systems "replicating or significantly improving themselves without explicit human authorization," and which last year's Nobel assembly on nuclear war never mentioned. This declaration puts it in harder terms than either.