Anthropic essentially bans OpenClaw from Claude by making subscribers pay extra

Foto: The Verge AI
As of Saturday, April 4, at 12:00 PM PT sharp, paid Claude subscribers are losing the ability to use their limits in third-party tools, such as the popular OpenClaw agent. Anthropic is officially cutting off free access via so-called third-party harnesses, forcing subscribers to transition to a pay-as-you-go model. This means that intensive AI usage in external interfaces will require additional payment, regardless of the existing subscription. The decision impacts the ecosystem of tools that gained recognition for automating email inbox and calendar management. Anthropic representatives, including Boris Cherny, explain the changes as a necessity to protect infrastructure resources, which were not designed for the specific, heavy traffic patterns generated by external applications. Although the company is offering a one-time credit equal to the value of a monthly subscription and discounts on data packages, the move is being interpreted as an attempt to more tightly bind users to its own products, such as Claude Cowork. For creative professionals and developers, this change signifies an increase in operating costs and the need to revise workflows based on external agents. The global trend of AI giants closing their ecosystems is becoming a reality—the priority is now API monetization and direct control over how and where model computing power is utilized. Users must now choose between the convenience of integrated tools and higher bills for the freedom to use open-source solutions.
The decision made by Anthropic on Friday evening caused an immediate stir in the community of advanced AI tool users. The company announced drastic changes to its subscription policy, which effectively cut off Claude users from free use of external wrappers and agents, such as the extremely popular OpenClaw. Starting from April 4, 2026, at 3:00 PM ET (12:00 PM PT), the limits included in the standard Claude subscription will no longer cover queries generated by third-party tools.
This move is seen as an attempt to regain control over infrastructure that has begun to show signs of overload under the pressure of automated agents. OpenClaw, which gained recognition for its ability to autonomously manage email inboxes, calendars, and even check in passengers for flights, generates traffic patterns that deviate significantly from typical human interaction with a chatbot. For Anthropic, a company building its business model on stability and sustainable growth, such intensive resource utilization by external "harnesses" has become unacceptable within current subscription plans.
The end of the "all-in-one" era for subscribers
The previous freedom, which allowed Claude subscription limits to be used in any compatible tool, is becoming a thing of the past. Boris Cherny, the director responsible for Claude Code at Anthropic, explained that subscriptions were simply not designed for the usage patterns characteristic of tools like OpenClaw. The company is now shifting to a pay-as-you-go model, meaning that every query sent by an external agent will be billed separately, regardless of the paid subscription.
Read also

Users who wish to stick with their current tools have two paths to choose from: using a Claude API key or purchasing new "usage bundles," which are to be offered at a discount. To mitigate the impact of this sudden change, Anthropic is granting subscribers a one-time credit equal to the monthly cost of their plan. For those completely dissatisfied with the new policy, there is an option to apply for a full subscription refund via a link sent in emails.
- Effective Date: April 4, 2026, at 3:00 PM ET.
- Main Change: Prohibition of using Claude subscription limits in third-party tools.
- New Options: Pay-as-you-go system, usage bundles, or API access.
- Compensation: One-time credit equal to the value of a monthly subscription.
Politics, competition, and personnel transfers
Industry analysts point out that behind the technical justification of "capacity management" may lie a broader competitive strategy. The creator of OpenClaw, Peter Steinberger, is currently an employee of OpenAI — Anthropic's biggest rival. Steinberger publicly admitted that along with OpenClaw board member Dave Morin, they tried to convince the Anthropic board to change its mind, but the only success was delaying the implementation of the new restrictions by just one week.

Restricting access for OpenClaw may also be a subtle signal encouraging users to switch to the company's native solutions, such as Claude Cowork. Promoting its own ecosystem allows Anthropic not only for better monetization but, above all, for full control over how their models are used in autonomous tasks. In an era of fighting for every teraflop of computing power, prioritizing customers who use the company's products directly or the official API becomes a business necessity.
"We want to be intentional in managing our growth so that we can continue to serve our customers sustainably in the long term. This change is a step in that direction." — Boris Cherny, Anthropic.
A new standard for AI agent monetization
The situation with Anthropic and OpenClaw sheds light on a broader problem in the AI industry: who should pay for "heavy" automation? AI agents, which perform hundreds of queries in the background to complete a single user task, consume resources at a rate that traditional "all-you-can-eat" subscription models cannot sustain. Anthropic is one of the first players to openly admit that free (within a subscription) use of such tools is economically unsustainable.
From the perspective of the Pixelift editorial team, this move sets a new standard in the relationship between model providers (LLM providers) and external software developers. The time of "guerrilla" tactics, where for $20-30 a month one could power massive automation engines, is coming to an end. The introduction of the pay-as-you-go system for tools like OpenClaw is a clear signal that professional automation will cost significantly more than standard access to a chatbot. It can be expected that other giants, like Google or OpenAI, will soon follow the same path, introducing a strict distinction between human and machine interaction.
For end users, this means the necessity to recalculate the profitability of their workflows. While the one-time credit and discounts on data bundles are a nice gesture, in the long run, using Claude as the brain for external agents is becoming a luxury reserved for those ready to pay for actual token consumption, rather than just the promise of unlimited access.









