This week the CEOs of OpenAI, Anthropic and Google DeepMind sat down with the leaders of the G7 – including Donald Trump, Keir Starmer and Emmanuel Macron – at a summit in Évian-les-Bains.
It was the first time all three had been in the same room at the same time for a meeting of this kind. Sam Altman used the moment to say something that sounded humble and probably wasn’t: “Do not cede your responsibilities to AI labs like mine. We develop the technology, and the citizens of the free world make the rules.”
On one level, that’s a reasonable position. On another, it’s the people who built the thing asking someone else to be responsible for it, while also warning that the thing might become the dominant source of military and economic power on Earth within two years. It’s a bit like handing someone a lit firework and saying, “how you handle this is really up to you.”
What They Actually Said, And Why It’s Complicated
The warnings delivered by the three CEOs were far from vague gestures. Instead, they offered stark, granular predictions that bypassed the standard discourse on AI.
Dario Amodei of Anthropic told G7 leaders he expects AI models to surpass human capability in most domains within one to two years. Demis Hassabis of Google DeepMind put the timeline at three to five years. Altman said he expects systems with “astonishing power” within another year or two. The specific concerns raised were national security ones – cyber threats, bioterrorism, nuclear capability, warfare – not the gentler anxieties about jobs or inequality that tend to dominate public conversation about AI.
These are people who understand this technology better than anyone. When they fly to the French Alps to warn world leaders about it, that warrants immediate concern. The complication is what came next.
When asked how governments should actually govern this technology, the three couldn’t agree. Amodei’s position is that the US and its democratic allies should form a coalition to set standards and isolate adversaries like China. Altman argued for broad access and erring toward human liberty, warning against concentrating too much power in too few hands – including, presumably, his. Hassabis called for an international technical standards body supported by leading labs, with standards updated quarterly as risks evolve.
Three people in the same room, building comparable technology, with fundamentally different views on who should control it and how.
Why The Real Issue Remains Unnamed
While the differences in opinion among these leaders are revealing, the deeper problem is structural. It’s a friction that has persisted across decades of technological shifts.
Democratic governance systems were not designed to regulate technology that moves at this speed. Legislation takes years and international frameworks take decades – the pace of AI advancement is measured in months. This is an inherent mismatch that has existed through every generation of transformative technology, from the internet to social media to genomics, and it has never been resolved. The same governments that are still arguing about social media regulation are now being asked to set standards for systems their advisors describe as potentially decisive for military and economic dominance.
The surrounding context highlights the irony. Days before the summit, the Trump administration enacted export controls that effectively crippled Anthropic’s global operations, offering no security details in return. Amodei arrived at the G7 advocating for clear rules, while simultaneously signalling a need for immunity from such administrative volatility. It was the most glaring, yet unspoken, contradiction.
Keir Starmer, asked about the best framework for international AI governance, talked about public consent. Macron urged democracies to cooperate, but no specific framework emerged from the summit.
So Where Does That Leave Everyone Else?
The clearest thing to emerge from the G7 summit is also the most unsettling: there is currently no consensus on how to govern AI, a blind spot shared by lawmakers and developers alike.
The incompatibility between how governments function and how technology evolves isn’t something one summit or framework can resolve. It’s a tension that has defined the relationship between private tech development and public governance for decades, and it’s now operating at a level where the stakes are measurably higher than before.
Altman is right to argue that responsibility shouldn’t rest with the labs. But he ignores the central difficulty: how can governments govern what they don’t fully understand, at a pace they aren’t equipped to match, when the developers themselves are in disagreement? The G7 didn’t provide the answers we need, and it’s doubtful the next summit will, either. However, a crucial shift has occurred: the conversation has moved past performative optimism toward an uncomfortable, necessary acknowledgement of the problem’s full, daunting scope.
