Anthropic Tightens Billing

Anthropic has done two things in a week that look separate only if you squint. First, it changed how Claude Code billing works for outside tools. Then it created a federal political action committee. One move is about compute. The other is about power. They belong in the same story. The billing change landed on April 4. Anthropic said Claude subscription plans were no longer meant to cover usage through third-party tools such as OpenClaw, a popular external agent framework. Users can still connect those tools to Claude, but now they have to turn on pay-as-you-go billing or bring a direct API key instead of leaning on a flat monthly subscription. Anthropic’s own pricing page still sells Pro at $20 a month and Max tiers at $100 and $200, with Claude Code included. What changed is the boundary around that inclusion. (tech.yahoo.com) That boundary matters because developers had found a cheap way to run expensive workloads. A consumer Claude subscription was being used, in practice, like a developer API plan. Anthropic’s explanation was blunt. Its subscriptions, the company said, were not built for the usage patterns of third-party tools, which were putting an “outsized strain” on its systems. The company said it was prioritizing people using Anthropic’s own products and API. That is not a minor policy cleanup. It is Anthropic closing a pricing loophole once the loophole became large enough to hurt. (tech.yahoo.com) The details make the shift clearer. Anthropic’s official Claude plans still present Claude Code as part of the subscription bundle. Its help center also promotes remote connectors and says Claude Code can work across paid plans. But the company has now drawn a harder line between first-party use and external harnesses that automate Claude at much higher volume. The practical message to power users is simple: if you want autonomous agents running through third-party software, Anthropic wants that metered like infrastructure, not bundled like a consumer app. (claude.com) That would already be a meaningful story on its own. But Anthropic paired it, almost simultaneously, with a move into electoral politics. Federal Election Commission records show that on April 3 the company filed to create the Anthropic PBC Political Action Committee, or AnthroPAC. The filing lists it as an active corporation PAC and a lobbyist or registrant PAC, tied directly to Anthropic PBC. The committee was registered one day before the billing crackdown became public. (fec.gov) A PAC is not just a press-release accessory. It is a machine for channeling money into relationships. Reporting this week says AnthroPAC will be funded by voluntary employee contributions, is expected to give to candidates in both parties, and is being set up as AI policy becomes a bigger fight in Washington. That places Anthropic in the same lane as the big tech firms it once seemed eager to distinguish itself from. The company that built its brand on safety language is now building the standard apparatus for influence too. (techcrunch.com) Seen together, the two moves are less surprising than they first appear. Anthropic is trying to control who gets scarce model capacity and on what terms. At the same time, it is investing in the political system that will shape procurement, regulation, and the legal boundaries of AI deployment. One step tightens the meter. The other reaches for the rulebook. The concrete paperwork is already there: AnthroPAC, committee ID C00946111, registered April 3, 2026, at Anthropic’s 548 Market Street address in San Francisco. (docquery.fec.gov)