National Security Agency is using Anthropic's Mythos despite blacklist, Axios reports

Why would they? This is the best advertisement they could have asked for, “people at top government institutions believe our product is that much better than the competition that they’ll use it even though it’s blacklisted”

Scuttlebutt is that Anthropic can't afford it. They made a really bad gamble buying committed compute last year and are have been getting dogfucked since late November by that decision. They're forced to buy compute at spot rates, and boy oh boy there just really isn't that much capacity left now that all the hyperscalers have their money fully in the mix. I won't be shocked if MS pivots to starve Anthropic out of compute (or loss-leader them with Azure credits like OpenAI) and absorb them. That would leave them with a functional monopoly on big commercial foundation models. The last 6-8 weeks have been marked by model output quality tanking, it's system prompts are adjusted frequently in an attempt to save them money but still have an LLM worth using. Opus 4.7 released last week, it _magically_ uses 20-30% more tokens, subscribers get a temporary credit, but once that expires, you've lost 20-30% of your on-paper capacity. Opus 4.7 is very aggressive in sub-models to piecemeal tasks to better-suited (read: better-suited _for Anthropic_, ie less expensive) ones, but it looks like they didn't let that bake quite long enough. Saw all the "success" of GPT-5's router and decided "yeah, let's do that."

Blacklist means nothing when you CTRL-F to delete any reference of the word black. So it just becomes a list.

Lmao, you think the NSA is going to tell everyone how they're using any of their tools, let alone Mythos?

Do you... know what slop is?

> They don't even say how they are using it You don't know what Mythos is, do you?