Anthropic has officially blocked third-party AI agents from accessing its Claude subscription levels. This change means that people can't use their flat-rate subscription limits on automation platforms outside of the site, like OpenClaw. The main reason for the crackdown is problems with managing resources.

Anthropic is giving users a one-time credit equal to their monthly subscription fee (good until April 17) and up to 30% off on usage bundles they already bought. The change in policy has quickly made developers angry. Many power users have said that switching from flat-rate subscriptions to metered billing makes things a lot more expensive. Now, standard agency workflows cost between $0.50 and $2.00 for each interaction, which makes it hard for solo developers and hobbyists to compete.

This change highlights a growing conflict in the industry: AI vendors need to tightly control and make money from their vast infrastructure, while developer communities keep pushing for cheap, unlimited access to the newest models. The company is putting the most important thing for customers who use its core products first: compute capacity. In the next few weeks, the restriction will apply to all third-party harnesses as well.

The new policy makes it clear that OAuth authentication is only allowed on the Claude Code and the official Claude.ai web interface. Developers use Claude's standard API key to log in and pay based on the usual metered pricing.