Anthropic announced in May 2026 that Claude paid subscriptions can once again power third-party AI agents — including OpenClaw — through a newly-introduced Agent SDK credits tier. The policy reversal undoes the April 2026 ban that had locked OpenClaw users out of subsidized Claude usage.
The Timeline
Early April 2026: Anthropic introduced policy language expressly prohibiting Claude Pro and Max subscriptions from powering non-Anthropic agents and harnesses. The stated reason: third-party agent loops bypass the response caching that makes flat-rate subscriptions economically viable, leading to runaway capacity costs.
April 4, 2026: The ban took effect. OpenClaw users running on Claude subscription quotas saw their requests blocked or rate-limited, depending on harness configuration.
May 2026: Anthropic introduced the Agent SDK credits subcategory, restoring third-party agent access on top of the existing chat allotment, but as a separate metered pool. The plan thresholds:
- Claude Pro ($20/mo): $20 of Agent SDK credits
- Claude Max 5x ($100/mo): $100 of Agent SDK credits
- Claude Max 20x ($400/mo): $400 of Agent SDK credits
What It Means for OpenClaw Users
OpenClaw and similar third-party agents can again use a Claude subscription, but with a credit pool that's separate from the conversational quota. The pricing structure is mostly “dollar-for-dollar” on the agent side — you're effectively paying API rates for programmatic use, with chat-style conversation continuing to flow through the higher-cap subscription pool.
For light agent users this is more than enough headroom. For heavy automated workflows the credit pool will run out faster than the conversational allotment, and users will need to either top up or fall back to direct API billing.
Why Anthropic Reversed
The April ban triggered immediate community pushback. Power users, open-source maintainers, and indie developers argued that locking out programmatic Claude usage damaged the ecosystem that had been driving Claude adoption in the first place. The Agent SDK credits structure is Anthropic's compromise: keep flat-rate subscriptions sustainable on the conversational side, while letting third-party agents run on metered credits that don't threaten capacity planning.
For OpenClaw Launch Users
If you're running OpenClaw Launch with a Claude provider connection, the new credit pool applies automatically — your subscription powers your agent on the Agent SDK side. OpenClaw Launch also continues to support OpenRouter and direct Anthropic API keys if you'd rather not depend on Claude subscription credits for agent work.