What Actually Happened
On April 3, 2026, Boris Cherny, Anthropic's head of Claude Code, posted a short announcement on X that set off a wave of responses from OpenClaw users around the world: "Starting tomorrow at 12pm PT, Claude subscriptions will no longer cover usage on third-party tools like OpenClaw."
By the following afternoon, the change was live. Claude Pro and Claude Max subscribers who had been routing OpenClaw through their subscription login found their agents were no longer authorized to use Claude's AI models within the flat-rate plan.
This was not a technical failure. It was a deliberate policy decision by Anthropic, and it had been building for a while.
For context: OpenClaw is the open-source AI agent platform created by developer Peter Steinberger and later acquired by OpenAI. It lets users build personal AI assistants that can browse the web, manage files, send emails, and do dozens of other tasks automatically. It is the fastest-growing open-source project in GitHub history, with hundreds of thousands of users worldwide.
Until this week, many of those users had been running OpenClaw by logging in with their Anthropic Claude account. The same OAuth authentication that lets you log into Claude Code also let OpenClaw access Claude's models. Users with a flat monthly subscription to Claude Pro (around $20 per month) or Claude Max (around $100 per month) could run as much OpenClaw activity as they wanted against those subscription limits.
Anthropic says that arrangement was never intended to work that way, and that the usage patterns created by autonomous AI agents running continuously were creating serious infrastructure strain.
Why Anthropic Made This Change
The short version: OpenClaw uses Claude very differently from a person typing questions into a chat window, and the economics did not add up for Anthropic.
A typical Claude Pro subscriber might send a few dozen messages a day. An OpenClaw instance running automated tasks can send thousands of requests, often continuously and without breaks. Multiply that by tens of thousands of heavy OpenClaw users operating on flat-rate plans, and you have a significant capacity problem for a company trying to serve millions of paying customers.
Anthropic's spokesperson told Business Insider that this usage "places an outsized strain on our systems" and technically violates the company's consumer terms of service. The terms have prohibited unauthorized third-party tool access since early 2024, but enforcement had been loose. Anthropic revised its terms more explicitly in February 2026, and last week's change was the formal enforcement.
Boris Cherny framed it plainly: "We've been working hard to meet the increase in demand for Claude, and our subscriptions weren't built for the usage patterns of these third-party tools. Capacity is a resource we manage thoughtfully and we are prioritizing our customers using our products and API."
The change also comes amid a broader wave of AI infrastructure pressure. Claude faced multiple outages in recent weeks as demand surged. Anthropic is managing a $38 billion company that raised $3 billion in Series G funding just this February, and it is clearly focused on ensuring its core products stay stable for the customers paying through official channels.
How OpenClaw's Creator Responded
Peter Steinberger, the creator of OpenClaw, was not quiet about his reaction. He posted on X: "woke up and my mentions are full of these. Both me and @davemorin tried to talk sense into Anthropic, best we managed was delaying this for a week. Funny how timings match up, first they copy some popular features into their closed harness, then they lock out open source."
Steinberger's comments touch on a tension that has been brewing in the AI developer community for months. Anthropic introduced several autonomous agent features in Claude Code, its own official AI agent product, earlier this year. Critics argue that locking third-party tools out of the subscription model conveniently clears the field for Anthropic's own competing products.
That said, Steinberger also gave credit where he felt it was due. In a follow-up post, he noted: "While I think what Anthropic does is sad for the ecosystem, I wanna give Boris credit for doing what he can to soften the fallout. Today's release will include some fixes for better cache use, to lower cost for API users."
That last detail matters for business owners. The OpenClaw team is actively working to reduce the API cost of running OpenClaw on metered pricing, so the transition to paid API access should become more affordable over time, not more expensive.
What This Means If You Run OpenClaw at Your Business
The impact depends entirely on how your OpenClaw setup was configured.
If your OpenClaw was connected via a Claude login (OAuth): Your agents stopped working at 12 PM Pacific on April 4. You need to reconfigure to one of the three options described below. Check your OpenClaw settings now and look for the authentication section.
If your OpenClaw was already using a Claude API key: You are not affected. API key users were never part of the subscription coverage arrangement, so nothing changes for you. Your billing model is already metered and compliant.
If your OpenClaw uses a different AI model (not Claude): You are not affected at all. OpenClaw supports multiple AI providers. If you configured it to run on GPT-4 or another model, Anthropic's decision does not touch you.
If you are evaluating OpenClaw for a future deployment: This changes your cost planning. The old approach of routing unlimited agent activity through a flat Claude subscription is no longer an option. You need to budget for metered API usage from the start. We cover that in the cost planning section below.
Your Three Options Going Forward
Anthropic has not left OpenClaw users with no path forward. There are three ways to continue using OpenClaw with Claude models, and each serves a different situation.
Option 1: Pay-As-You-Go Extra Usage Billing
Anthropic has added an "extra usage" add-on to existing Claude subscriptions. You keep your Claude Pro or Max plan and enable overage billing on top of it. When your agent activity exceeds your base plan limits, you pay per token rather than being blocked.
This is the simplest transition if you want to keep using your Claude login with OpenClaw. You do not need to set up API keys or change your authentication method. Anthropic says discounted usage bundles are available as part of the transition to help offset the initial cost jump.
This option works best if your team uses OpenClaw in bursts rather than continuously. For light-to-moderate agent usage, the base subscription plus occasional overage charges may still come out cheaper than full API pricing.
Option 2: Discounted Usage Bundles
Anthropic is offering one-time transition credits and pre-purchased usage bundles at a discount. If you received a notification email about this change (Anthropic is emailing affected subscribers directly), that email contains a link to purchase bundles at below-standard rates.
Check your email account associated with your Claude subscription if you have not already. Affected users also receive a one-time credit equal to their monthly plan cost. That credit can be applied toward API usage or usage bundles, effectively giving you a free month to figure out your long-term plan.
Option 3: Standard Claude API Access
The most robust long-term option is to switch OpenClaw to use a Claude API key rather than a subscription login. An API key connects OpenClaw directly to Anthropic's infrastructure on a metered, per-token basis. You pay for exactly what you use, with no flat-rate cap to worry about.
This is how Anthropic wants businesses to integrate Claude into their tools, and it is the arrangement that is explicitly supported long-term without risk of further policy changes. If you are running OpenClaw in a business environment with predictable workloads, API key access also lets you use the full range of Claude models, including the latest versions.
To set this up: create an API key in your Anthropic account dashboard, then update your OpenClaw configuration to use that key in the model settings. The OpenClaw documentation has a step-by-step guide for this. The OpenClaw team is also releasing an update with better token caching, which will lower your per-task cost.
How to Think About the Cost Change
The most common concern we are seeing from business owners is cost. Under the old arrangement, a Claude Max subscription at around $100 per month gave you effectively unlimited agent activity. Under API pricing, you pay per token, which can add up quickly for an agent running all day.
Here is a realistic way to think about it.
Most business OpenClaw deployments fall into one of two categories: assistant tasks (email drafting, research, summarization, scheduling support) and process automation (recurring data checks, file organization, workflow triggers). These use very different amounts of token capacity.
For assistant tasks where an employee interacts with the agent a few dozen times a day, Claude's mid-range models at current API pricing typically cost $5 to $20 per month per user. That is comparable to or cheaper than a Claude Max subscription once you account for the fact that you are only paying for what you use.
For process automation with agents running continuously in the background, costs can be higher. The key is to choose the right Claude model for each task. You do not need Claude's most powerful model to check whether a form was filled out correctly or to summarize a daily report. Anthropic offers a range of models at very different price points, and OpenClaw lets you specify which model each type of task uses.
The OpenClaw team's update to improve token caching efficiency will also help. Caching means repeated context (like your agent's instructions and permissions) does not get re-billed on every request, which can cut costs significantly for long-running agents.
If you are not sure how to estimate your usage, Anthropic's API dashboard shows a usage log once you start making API calls. Run your typical workload for a few days and let the numbers tell you what to expect monthly.
What This Means for NemoClaw Users
If your business runs NemoClaw, Nvidia's security-hardened enterprise version of OpenClaw, the same authentication changes apply. NemoClaw sits on top of OpenClaw's foundation, so the model connection works the same way.
That said, NemoClaw deployments are typically configured by IT teams who set up API key access from the start, because the enterprise setup process requires explicit API credentials rather than a personal login. If your NemoClaw deployment was set up correctly by your IT team, there is a good chance you are already using an API key and are not affected.
Check with whoever manages your NemoClaw environment. Ask them specifically whether the AI model connection uses OAuth (a login) or an API key. If they are not sure, that is a sign the setup needs a review.
For organizations considering NemoClaw for the first time, this change actually makes the case for a proper enterprise deployment stronger. An IT-managed configuration with explicit API keys and defined usage budgets is exactly what this situation calls for. You get control over costs, clear audit trails, and no exposure to consumer account policy changes. The NemoClaw for business guide walks through what that setup looks like.
The Bigger Picture: AI Platforms and Platform Risk
Beyond the immediate action items, this situation is worth thinking about as a business owner in a broader sense. It illustrates what the technology industry calls "platform risk."
Platform risk means your tool depends on another company's decisions. Anthropic changed a policy and, with roughly 24 hours of notice, thousands of businesses woke up to broken workflows. Even though Anthropic provided transition credits and alternatives, the disruption was real and the timing was not the affected users' choice.
This is not unique to Anthropic or to OpenClaw. It is a pattern that plays out across every technology platform. App developers on Apple's App Store have faced sudden rule changes. Businesses built on social media APIs have had access revoked overnight. Cloud services have deprecated features without warning.
The lesson is not to avoid AI tools. The lesson is to structure your AI deployments with enough flexibility to handle changes. A few practices that reduce platform risk:
Use API keys instead of personal logins for business-critical tools. API access is designed for programmatic use and is governed by different terms than consumer subscriptions. In this case, API key users were not affected at all.
Test whether your tool can run on more than one AI model. OpenClaw supports multiple providers. A deployment that works with both Claude and another capable model means one provider's policy change does not take your operation offline.
Keep your AI costs in your business accounts, not your personal ones. When AI tools are tied to personal subscriptions, policy changes hit personally. API keys connected to business accounts give you proper cost tracking and clear ownership.
Watch what your AI vendors announce. Anthropic updated its terms in February 2026, two months before enforcement. If your team had caught that change, you would have had time to plan the transition on your schedule rather than scrambling after the fact.
The broader trend here is also worth noting. As AI companies see their infrastructure costs grow with demand, the era of "all you can use for $20 a month" for heavyweight automation is ending. The economics do not work when AI agents run continuously at the same flat rate as occasional human users. Business owners who factor this into their planning now will be better positioned than those who get surprised by the next wave of policy changes.
What OpenAI (OpenClaw's Owner) Has Said
OpenAI acquired OpenClaw and its creator Peter Steinberger in early 2026 for a reported sum in the multi-million dollar range. That means OpenClaw is now an OpenAI-owned product, while Claude is an Anthropic product. The two companies are direct competitors.
This context matters because it shapes the likely long-term trajectory. OpenAI has strong incentives to make OpenClaw work well with its own models (GPT-4o, o3, and future releases) and less incentive to smooth over the friction Anthropic just created. If you have been running OpenClaw on Claude primarily because Claude is excellent, now is a good time to test whether GPT-4o or another model meets your needs for the tasks you actually use the agent for.
Semafor reported this week that Anthropic is exploring building its own version of OpenClaw, which would compete directly with the product it just cut off. Whether or not that materializes, the competitive dynamic between these two companies is clearly shaping product and policy decisions in ways that affect business users caught in the middle.
The practical upshot: if your workflows depend on Claude's specific capabilities (and some legitimately do, especially for nuanced writing or reasoning tasks), the API key route is the right long-term answer. If you are flexible about which AI model powers your agent, this is a natural moment to test alternatives and reduce your dependence on any single provider.
What to Do Right Now
Here is a plain-English action checklist for business owners and IT managers.
Step 1: Check how OpenClaw is connected to Claude. Open your OpenClaw settings and look at the model configuration. If it shows a Claude login or OAuth connection, you are affected. If it shows an API key, you are not.
Step 2: Check your email. Anthropic is sending transition information to all affected subscribers. Look for an email from Anthropic about the third-party tool change. It contains a link for the transition credit and information about purchasing discounted usage bundles.
Step 3: Choose your path. Review the three options described earlier in this article. For most businesses, Option 3 (API key access) is the cleanest long-term solution. For light casual users, the extra usage billing add-on may be simpler.
Step 4: Update your OpenClaw configuration. If switching to an API key, create one in your Anthropic account dashboard under the API section, then enter it in OpenClaw's model settings. Test with a simple task to confirm the connection is working before relying on it for business workflows.
Step 5: Review which model you actually need. When switching to API pricing, take the opportunity to check whether each of your agent tasks needs Claude's most powerful model or whether a lighter, cheaper model works just as well. The savings can be significant.
Step 6: Update your budget estimate. Monitor API usage in your Anthropic dashboard for the first week after switching. You will get a realistic picture of what your agent activity actually costs per month under metered pricing.
The Bottom Line
Anthropic's decision to block OpenClaw from Claude subscriptions is disruptive but not a disaster. The tools still work. The path forward is clear. The cost of running OpenClaw on Claude via API key access is manageable, especially with the efficiency improvements the OpenClaw team is shipping alongside this change.
What this moment really highlights is the importance of treating AI tools like the business infrastructure they have become. Set them up properly with API keys and business accounts. Monitor the terms of service for the platforms you depend on. Build in some flexibility so that one company's policy change does not take your operation offline overnight.
OpenClaw remains one of the most capable AI agent platforms available. NemoClaw adds the enterprise security layer that business deployments need. Neither product is going anywhere. The connection to Claude just got more formal and more metered, which is actually how business software relationships are supposed to work.
If you are earlier in your OpenClaw journey and want to understand what the platform does and whether it belongs in your business, the What Is OpenClaw guide is the right starting point. If you are specifically thinking about how to deploy it securely in an organization, start with the NemoClaw overview to understand how Nvidia's enterprise layer addresses the security considerations that come with giving an AI agent access to your business systems.