AI vendors tighten control over who can use them

Swedish government offices are banning ChatGPT and Claude in favor of Microsoft Copilot, which commenters read as a contractual side effect of Microsoft licensing rather than a safety judgment. In a related thread, Anthropic stopped letting Claude Code subscriptions run through OpenClaw, citing 'outsized strain on our systems.' Users answered that they were using the capacity they paid for. Xbox's plan to charge 24 to 40 cents an hour to stream games you already own sits in the same family of moves.

The pattern is that the subscription deal changes after you build a workflow on it. Enterprise bundling wins over best-model choice, and third-party harnesses get cut when they use too much. Commenters on the OpenAI 'why not start 100 companies' thread make the same point from the other side: selling tokens is 'risk-free money,' so vendors have every reason to protect that channel.

Counterpoint: some thread voices expect agent-agnostic tooling to spread precisely because of these lockouts.