Anthropic Policy Fights: Pentagon Ban Reversed, Claude Restrictions Tightened
A US judge ruled that the Pentagon's blacklisting of Anthropic was unlawful, clearing the way for Anthropic to pursue government contracts and potentially collect damages for lost revenue during the ban. The ruling was described as straightforward, with the government's evidence called essentially nonsensical by commenters reading the opinion.
At the same time, Anthropic is tightening the screws on how Claude can be accessed. The OpenClaw thread (third-party Claude Code harness) shows Anthropic cutting off subscribers who use unofficial tooling, citing system strain. This frustrated paying subscribers who argued they were using exactly what they paid for in tokens. The underlying tension: Anthropic wants to monetize agentic use through its own API pricing, not through flat subscriptions being used at maximum intensity through third-party wrappers.
These two stories together show Anthropic operating simultaneously at the policy frontier and the product frontier, winning in court while also frustrating power users. Neither is obviously the right call.
So what?
If your product depends on Claude via subscription-tier access through unofficial tooling, treat that as a temporary arrangement, not a business model. Anthropic has made clear that heavy agentic use belongs on the API with metered pricing. Budget accordingly, or build model-agnostic pipelines that can shift to alternatives when policies change.