A Contradictory Picture
The tension in this news is its self-contradiction: on one side the Pentagon's confrontational posture of putting Anthropic on a blacklist, on the other the intelligence system continuing to procure and use its model. The explanation isn't actually complicated—the US government isn't a monolith, the Defense Department's supplier policy doesn't govern the intelligence apparatus's procurement channels, and the NSA's selection logic is pragmatic: use whichever model works best, and political posturing is another department's business. Mythos is said to be Anthropic's model for high-security scenarios, fitting exactly this need.
The Entanglement of AI Companies and Government
Placed within the whole tug-of-war saga between Anthropic and the Pentagon, this is more interesting: bans, lawsuits, exemptions, continued use—the plot keeps reversing, showing neither side has a real alternative. The government can't do without frontier models, and the frontier labs can't afford to lose government orders, so the fight over safety review and usage policy is essentially bargaining. The long-term impact on the industry is in the precedent: how the relationship between AI suppliers and the national-security machine is defined, and how much right a model company retains to say no to military uses—these boundaries are precisely being drawn in messy individual cases like this.
via: Hacker News