A thread on Hacker News AI Best is currently overflowing, and the discussion revolves around something many in the AI community have been closely following since early April: Anthropic reversed its stance on third-party tools like OpenClaw just over two weeks ago, and now OpenClaw documentation reports that things have been cleared up – use via Anthropic is permitted again.
For those who haven't followed this drama: On April 4, 2026, Anthropic fully enforced a ban on using Claude Pro/Max subscriptions as the engine for unauthorized third-party tools. The reasoning was straightforward – these tools placed "disproportionate pressure" on the infrastructure, and Anthropic wanted to prioritize capacity for its own products and official API use. Boris Cherny, who leads Claude Code at Anthropic, was quite clear on this.
OpenClaw is a rapidly growing open-source project created by Austrian Peter Steinberger – a self-hosted AI agent that communicates with you through WhatsApp, Telegram, iMessage, and a host of other messaging platforms. It is precisely the type of tool Anthropic targeted with the ban.
The most likely answer, based on what the HN comment section is unearthing, is that this is not a full reversal. The distinction Anthropic has always operated with – API keys are fine, OAuth tokens from subscriptions are not – remains. What OpenClaw has likely done is adapt: the documentation now reflects that the tool supports genuine API key authentication, not subscription tokens. In other words: you can still use Claude through OpenClaw, but you pay per token via API, not through your flat-rate subscription.
Why is this interesting? Because it illustrates a pattern we will see much more of: AI providers are tightening control over how their models are used, and open-source projects must adapt or die. Anthropic offered affected users a one-time credit equivalent to the monthly price (redemption deadline April 17, already expired), and launched "Extra Usage" packages with up to 30% discount for pre-purchase.
This is still an early signal from community sources, and the actual scope of what OpenClaw now permits has not been directly confirmed by Anthropic. But with 498 points on HN in a relatively short time, this is something mainstream tech press will pick up shortly. Stay tuned.