1. If AI stalls out, you're left with an extremely centralised panopticon, and yes I think this would be much worse for space of possible outcomes vs status quo. The "plan" in Plan A is to make enough space and give enough money for *someone else* to figure out alignment in an unspecified fashion, while we have a massive enforcement checkpoint and bureaucracy. 3. State capacity isn't scalar. USG can quite easily be good at export controls and bad at tech evals. Orthogonal to just importance. "I agree the state capacity issue is rough, but sometimes to do good stuff, you need capacity. And it doesn't seem impossible if USG thinks of AI as by far the most important issue." It would be nice if the world worked like this. Look around you, does this seem like the world we're living in? This is Panglossian. 4. "Governments require companies to write up detailed arguments for why their new AIs won’t cause an irreversible catastrophe." This is going to end up in essay writing or license raj. If you operationalise "Reasonable assurance" I might be more persuaded here. I'm not sure what we're learning even today from the existing model cards beyond building the muscle to write model cards - Grok 4.5 doesn't have one for instance. 5. I hadn't seen your supplement on TRT, its better defined, but note it destroys the labs IP. Which I think its unworkable, extreme, and dangerous, and will only serve to kill AI industry or put it in a sort of soviet/china style grip. I don't get why this would be desirable if the goal is to have competition to get more frontier models and/or especially risk from RSI if it's even more centralised? 6. I did read the Plan A assumption appendix. It does not answer the questions I raised. It is mostly concerned with some sensitivity measurements within your model, not about the model. Right now you state a technical assumed problem, pose a way to get time/moeny to solve the assume technical problem with a solution, and create a whole bunch of political problems alongside which aren't addressed. 7. A specific concrete example might be "here's the treaty we should sign with china" then work through the possible consequences and defections. Red team it. Then it's a specific proposal we can do actually discuss! I think in general you do high levels of rigour within your own model, low rigour outside the model or about the model, and very low rigour to political considerations of what would happen even at step one of this plan.

It is not decentralised, how will this be enforced? And why would many people enter the indsutry? It'd be like more people don't enter utilities. I didnt mention China but if code's free they win (like in every buildout we know of). Re the last more wargames might be an easy first step. But yes it's the biggest gap today.