This isn't too surprising to a point. It will be interesting to see the level that they allow Anthropic back into the fold as they will clearly not fully let it back, especially within the Department of War (Defense) setting. The biggest driver will again be the extent to which Anthropic requires/demands the input dataand the results.

If they require all of the inputs and outputs go to them then the ability to use them for any sort of serious work in the government would be limited. After all, you wouldn't want a single company to start to exclusively control the government's own data.

Anthropic played the Mythos card very, very well, and I think it is pretty clear they knew they had this card when they tried to nuke the DOW. Once they suffered the huge blow-up and fallout associated with it, they had to prove that this model had the capabilities they believed it did, and apparently, they were not lying.

I will be interested to see if this affects what people think of Anthropic. They gained a ton of users for getting kicked out of the U.S. Government, so if they go back, will people who dumped OpenAI for Claude change their tunes?