> And despite all this they've still held their two red lines

How do you know? What reason is there to trust a company like Anthropic? It's about money, a lot of money, and from a company's perspective there is no reason to stick to these claims while the competition doesn't care and can train and improve their models on the vast pile of data they receive while mass surveilling U.S. citizens (which clearly is a huge competitive advantage).

[dead]

With the IPO on the horizon, Anthropic has to show that their models are significantly improving and are far better than other models from the competition while concealing that their business is economically unviable. They solve this conflict by throwing even more hardware resources at their models to upscale their "emerging capabilities" even more. At the same time, this effectively makes their offered services economically even more unviable. So they have to put in some hard limits, which they do by implementing "security guardrails", which incidentally usually kick in in non-security contexts.

Anthropic is primarily burning U.S. based capital and investments, from the administration's standpoint it's simply rational to stop letting non-U.S. actors burn this huge pile of U.S. capital. Of course, it's also a direct support of Anthropic by the administration (they are from the same social class) for their IPO: "These models have huge capabilities and are dangerous, we have to limit access", so that potential investors may conclude that an investment will surely lead to huge gains in the future. Anthropic is really good at marketing after all.