The slowdown pact met its first government on Monday, and the government said no. Germany’s digital affairs ministry, responding to calls from top AI developers to pace the frontier, stated flatly that halting AI development is not a viable option for Europe — continued innovation is needed for digital sovereignty, and any governance regime that actually matters will require the involvement of the United States and China, neither of which signed anything. Berlin is taking the position to the G7. The most consequential sentence in the reply is the quiet one: the ministry concedes the risks are real, including systems that could escape test environments and reach other systems on their own — an “entirely new threat scenario,” in its words, that demands a new policy response.
Read the reply against the weekend it answers. On Friday the field’s four biggest names converged on a single sentence — pace capability until safety catches up — and markets repriced the capex supercycle on the news. The assumption underneath the pact was that governments would welcome restraint from the companies they have spent two years trying to regulate. Germany’s answer is the first sovereign counter-read: welcome to our problem. Europe does not have a frontier lab to pace. What it has is a dependency — on American models, American accelerators, American cloud — and a ministry that has concluded that slowing down the people you depend on is not safety policy, it is exposure.
The sovereignty frame is doing visible work. This is the same government whose chancellor argued in the spring that industrial AI needs less stringent EU regulation, not more — the ministry’s Monday statement is that instinct, formalized and pointed at the G7. Notice what Berlin is not saying: not that the risks are exaggerated, not that the pact is theater. The statement accepts the threat model almost verbatim — escape from test environments, autonomous reach into other systems — and then declines the remedy on structural grounds. That is a harder position to argue with than denial, because it agrees about the fire and objects to who is holding the hose.
The Seat Problem
The demand for US and Chinese involvement is the statement’s real content, and its sharpest edge. A slowdown among four Western labs does not bind a single lab outside the pact — the trust problem the pact itself carries — and Berlin has now said the obvious part out loud: any regime that binds only the signatories is not governance, it is market shaping. Bringing Washington and Beijing to the table is the only version that arithmetically works, and it is also the version nobody knows how to convene. The G7 is where Germany wants the conversation; it is not obviously where China answers.
For the labs, the lesson lands uncomfortably. The pact’s authors wanted their restraint read as leadership — the industry putting its own governor on before regulators did. Berlin’s reply reframes it: from where a dependent economy sits, a voluntary slowdown by your suppliers looks less like leadership and more like supply management. The next moves are now visible in both directions — the signatories must either show verification that makes restraint legible, or watch the conversation migrate from safety to dependency. The brakes found company over the weekend. On Monday they found out that company has its own destination, and it is not the roadside.
The Takeaways
- Germany’s digital ministry says halting AI development is not viable for Europe — sovereignty requires continued innovation, and governance needs the US and China involved.
- The ministry accepts the labs’ threat model — including AI escaping test environments as an “entirely new threat scenario” — but declines the slowdown remedy on dependency grounds.
- Berlin is taking the position to the G7, reframing the slowdown pact from safety leadership to supply management for dependent economies.
