I feel like the concern underpinning this aged pretty well. UK and European nations can no longer rely on their relationships and agreements with US companies, because of the US government - especially as they move towards needing the most cutting-edge tech embedded in more critical processes.
I wonder the degree to which the USG has thought through the consequences of this.
"What we now know, as of last Friday at 5:01pm DC time, is that no agreement made with one of these companies can be relied on. No red lines mutually agreed on can be trusted. Yes, Dean (and Anton, and others) are right that the UK, Europe and others are at a massive and likely IMO insurmountable structural disadvantage to get to the frontier. But even having sub-frontier systems at this point may be preferable to this degree of vulnerability."
Aw hell, let's spell it out. If they're willing to do this to one of their own, because they hold the line on domestic surveillance - what does this mean for us 'allies' who partner with and use this tech?
There's the leverage over withholding the tech once it's embedded in our economies and essential to functioning economies. Play it forward. Next time they want to demand to overrule territorial sovereignty like Greenland? Next time we have airbases they want to use for a war we haven't agreed to be in?
And there's the surveillance. These tools will become very powerful in the coming years, Anthropic aren't wrong about that. And our economic, health, commercial and perhaps even security data will be in there.
What we now know, as of last Friday at 5:01pm DC time, is that no agreement made with one of these companies can be relied on. No red lines mutually agreed on can be trusted. Yes, Dean (and Anton, and others) are right that the UK, Europe and others are at a massive and likely IMO insurmountable structural disadvantage to get to the frontier. But even having sub-frontier systems at this point may be preferable to this degree of vulnerability.
And of course, the topic I worry most about: coordination around navigating safety and risk. Hard to know where even to start, but it's not good. I'll write more on this when I can.