Anthropic restricts Claude subscriptions — third‑party agents like OpenClaw blocked, users moved to pay‑as‑you‑go

Overview

Anthropic has moved to stop subscribers from running third‑party automation agents (notably OpenClaw) on Claude, it has been reported. From April 4, subscription tokens can no longer be used to run external tools; anyone who wants to keep using those agents must switch to pay‑as‑you‑go API billing. The change follows a series of quiet technical and contractual steps and, according to the company’s staff, reflects a deliberate policy shift to prioritise official product and API usage over heavily automated third‑party workloads.

Timeline and company posture

Boris Cherny, head of Claude Code, reportedly posted multiple tweets explaining the clampdown and stressing that subscriptions were never designed to subsidise continuous third‑party agents. It has been reported that Anthropic began closing the technical loopholes in January by restricting subscription OAuth tokens, updated its terms of service in February to ban subscription use by third‑party automation, and in March bolstered Claude Code to reclaim core capabilities from external tools. The company also reportedly offered affected subscribers a one‑time credit equivalent to a month’s fee; refunds and discounts were said to be on the table for some users.

Why now — costs and competition

Why the sudden hard line? Because agent tooling can massively amplify usage. It has been reported that running an OpenClaw‑style agent continuously can carry an API cost in the thousands of dollars per day — figures cited range from roughly $1,000 to $5,000 for a single persistent agent — quickly destroying the economics of a fixed‑price subscription. At the same time, competitors are taking different approaches: OpenAI is reported to have welcomed the opportunity, even hiring OpenClaw’s founder, and positions itself to absorb demand rather than restrict it. The result is not just a product decision but a strategic split in how providers monetise and ration compute.

Industry implications

The move underscores a broader reckoning in the AI market: subscriptions work until automated agents break the assumptions behind “average” usage. Providers now face a stark choice — raise subscription prices, restrict use, or lose money. For users and developer ecosystems, that means agent builders will chase new compute “exits,” and some workloads may migrate between providers. It has been reported that Anthropic chose to draw a line now rather than erode its business permanently. What happens next will depend on who can offer cheaper, more efficient inference at scale — and who can live with the economics of 24/7 automated agents.