4 Comments
User's avatar
Jacob's avatar

I think you underestimate institutional persistence. The AI Office was just set up and has sole authority on GPAI models, creating path dependency that's historically difficult to reverse.

For point 2, could you clarify whether you mean labs won't deploy in the EU due to the AI Act, or that they'll give model weights to specific EU countries through bilateral agreements? I believe these macro forces mostly affect the strictness of enforcement rather than institutional authority. Do you have any examples where tech regulation authority actually transferred back from the EU to member states?

Given current AI developments and timelines, I find it hard to understand why you describe such a scenario without clarifying that it's meant for 2+ years in the future. While member states will compete over AI resources, the COVID vaccine procurement example you mention actually contradicts your argument - it was handled centrally despite similar incentives for fragmentation.

Especially regarding Germany, I don't see evidence of willingness to contribute meaningfully to independent AI policy outside the EU framework. Do you have specific information suggesting otherwise?

I agree with Herbie Bradley's point that compliance costs for large AI labs may not be prohibitive enough to trigger your bilateral agreement scenario in the near term. The most likely outcome seems to be labs doing additional evaluation and reporting work, with approvals coming slightly later than US deployments.

Expand full comment
Anton Leicht's avatar

Thank you! I think I disagree with some of your characterisations.

First, the AIO only has authority insofar as it's provided by the AIA, there is no automatic mechanism by which the AIO gets do a lot out of its scope. And I think discussions around the CoP in the next weeks will provide a lot of interesting information on this, too, I'm curious what we'll think in a month or two.

Second, I think there are a lot of conceivable scenarios where deploying models, but especially hosting them at large scale, outside of the US becomes similarly constrained to compute today. We see some thinking along those lines at the intersection of NatSec / AI discussions today, and I think it's likely to expand in due time. That'll come with conditions, either outright quid-pro-quos or at least security stipulations etc, which might lead some member states to qualify and some not to (compare this to the diffusion framework, for instance).

Do I have examples of this happening so far? De-jure-examples, no (but the piece also doesn't project a clear de-jure change). De facto, there are a bunch of member states that the Commission does not enforce some regulation aspects with as stringently as possible for some countries, Ireland and GDPR is one of the most striking examples there.

Third, I agree - the vaccine example is the opposite case. But as we try to argue: It was really hard to pull off, and it exhausted a lot of capital for trying ther same thing again.

Fourth: Re Germany, I sadly can't comment too in-depth on it! But for some early signs, look at the CDU party platform on the AIA, at some early positioning on AI sovereignty, and at the extent of alignment and orientation regarding France. I suspect there will be a lot more quotable things on this soon-ish, though mostly after a member state 'wake-up' to AI progress, which could still take a year or two.

As to why not clarify timelines: I'm not so sure about how quickly or slowly that'll go; to my mind, that has a lot to do with capability trajectory, public reaction, and US foreign policy. I find these three areas a bit tricky to predict at the moment, so I feel it's hard to pin down how exactly this will unfold; just that the headwinds only get stronger from here. The rest, I think, will be left to see!

Expand full comment
Herbie Bradley's avatar

I agree with the general thrust of this, but:

> As this happens, access to powerful AI becomes increasingly important for national productive capacity and so EU member states will face mounting pressure to weaken the enforcement of the AI Act, and or make bilateral agreements with AI makers, to access the most advanced models.

given the political pressure on the EU Commission to go easy on US companies, and that it probably realistically doesn't take too much effort to comply for a large AI lab (unless releasing open-source) for the foreseeable future, then don't you think the likely outcome is just that labs do a little more evals and reporting logistics work than normal and get approved by the AI Office a month or so later than any US deployment? I'm not sure we should expect bilateral agreements under that scenario.

Expand full comment
Anton Leicht's avatar

Thanks! To my mind, bilateral agreements on economic access happens a bit later on & are mostly seperate from the current regulatory situation.

I think what you describe is a realistic model of what happens in the next months / 1-2 years. Later on, I could imagine that US 'AI exports' / clandestine API access / sharing models to be run on foreign clusters would be more constrained, e.g. by being leveraged as a foreign policy tool or by being associated with security provisions, right-of-accesss to US firms, etc. (not incomparable to existing diffusion rules around compute).

Given the current tone of the USG, I could also imagine that model access comes with some sort of 'free speech' or nonenforcement of EU rules commitment that has very little to do with how costly these rules actually are and much more with their political standing in the US; that's where this point interfaces most closely with the regulation part from before.

In either case, I'd think some member states would be in a much better position (due to existing security infrastructure, above-average bilateral relationship with US compared to other member states, more foreign policy concessions to trade against access..) than others to ensure their own continued access, and heavily motivated by economic incentives to do so, even if it meant not moving in lockstep with the rest of the EU.

Expand full comment