Anthropic has rolled out a new set of technical controls aimed at stopping third-party applications from impersonating its official coding client, Claude Code, to gain cheaper access and higher usage limits to Claude AI models. The move has directly disrupted workflows for users of popular open-source coding agents such as OpenCode.
At the same time—but through a separate enforcement action—Anthropic has also curtailed the use of its models by competing AI labs, including xAI, which accessed Claude through the Cursor integrated development environment. Together, these steps signal a tightening of Anthropic’s ecosystem as demand for Claude Code surges.
The anti-spoofing update was publicly clarified on Friday by Thariq Shihipar, a Member of Technical Staff at Anthropic working on Claude Code. Writing on X (formerly Twitter), Shihipar said the company had "tightened our safeguards against spoofing the Claude Code harness." He acknowledged that the rollout caused unintended side effects, explaining that some accounts were automatically banned after triggering abuse detection systems—an issue Anthropic says it is now reversing.
While those account bans were unintentional, the blocking of third-party integrations themselves appears to be deliberate.
Why Harnesses Were Targeted
The changes focus on so-called “harnesses”—software wrappers that control a user’s web-based Claude account via OAuth in order to automate coding workflows. Tools like OpenCode achieved this by spoofing the client identity and sending headers that made requests appear as if they were coming from Anthropic’s own command-line interface.
This effectively allowed developers to link flat-rate consumer subscriptions, such as Claude Pro or Max, with external automation tools—bypassing the intended limits of plans designed for human, chat-based use.
According to Shihipar, technical instability was a major motivator for the block. Unauthorized harnesses can introduce bugs and usage patterns that Anthropic cannot easily trace or debug. When failures occur in third-party wrappers like OpenCode or certain Cursor configurations, users often blame the model itself, which can erode trust in the platform.
The Cost Question and the “Buffet” Analogy
Developers, however, have largely framed the issue as an economic one. In extended discussions on Hacker News, users compared Claude’s consumer subscriptions to an all-you-can-eat buffet: Anthropic offers a flat monthly price—up to $200 for Max—but controls consumption speed through its official Claude Code tool.
Third-party harnesses remove those speed limits. Autonomous agents running inside tools like OpenCode can execute intensive loops—writing code, running tests, fixing errors—continuously and unattended, often overnight. At that scale, the same usage would be prohibitively expensive under per-token API pricing.
"In a month of Claude Code, it's easy to use so many LLM tokens that it would have cost you more than $1,000 if you'd paid via the API," wrote Hacker News user dfabulich.
By cutting off spoofed harnesses, Anthropic is effectively pushing heavy automation into two approved channels: its metered Commercial API, or Claude Code itself, where execution speed and environment constraints are fully controlled.
Community Reaction and Workarounds
The response from developers has been swift and mixed. Some criticized the move as hostile to users. "Seems very customer hostile," wrote Danish programmer David Heinemeier Hansson (DHH), creator of Ruby on Rails, in a post on X.
Others were more understanding. "anthropic crackdown on people abusing the subscription auth is the gentlest it could’ve been," wrote Artem K aka @banteg on X. "just a polite message instead of nuking your account or retroactively charging you at api prices."
The OpenCode team moved quickly, launching a new $200-per-month tier called OpenCode Black that reportedly routes usage through an enterprise API gateway rather than consumer OAuth. OpenCode creator Dax Raad also announced plans to work with Anthropic rival OpenAI so users could access Codex directly within OpenCode, punctuating the announcement with a Gladiator GIF captioned "Are you not entertained?"
The xAI and Cursor Enforcement
Running parallel to the technical crackdown, developers at Elon Musk’s AI lab xAI reportedly lost access to Claude models around the same time. While the timing suggested coordination, sources indicate this was a separate action rooted in Anthropic’s commercial terms.
As reported by tech journalist Kylie Robison of Core Memory, xAI staff had been using Claude models through the Cursor IDE to accelerate internal development. "Hi team, I believe many of you have already discovered that Anthropic models are not responding on Cursor," wrote xAI co-founder Tony Wu in an internal memo. "According to Cursor this is a new policy Anthropic is enforcing for all its major competitors."
Anthropic’s Commercial Terms of Service explicitly prohibit using its services to build or train competing AI systems. In this case, Cursor itself was not the issue; rather, xAI’s use of Claude through the IDE for competitive research triggered the block.
This is not the first time Anthropic has cut off access to protect its models. In August 2025, the company revoked OpenAI’s access to the Claude API under similar circumstances. At the time, an Anthropic spokesperson said, "Claude Code has become the go-to choice for coders everywhere, and so it was no surprise to learn OpenAI's own technical staff were also using our coding tools."
Earlier, in June 2025, the coding environment Windsurf was abruptly informed that Anthropic was cutting off most first-party capacity for Claude 3.x models. Windsurf was forced to pivot to a bring-your-own-key model and promote alternatives like Google’s Gemini.
Together with the xAI and OpenCode actions, these incidents underscore a consistent message: Anthropic will sever access when usage threatens its business model or competitive position.
Claude Code’s Rapid Rise
The timing of the crackdowns closely follows a dramatic surge in Claude Code’s popularity. Although released in early 2025, it remained niche until December 2025 and early January 2026, when community-driven experimentation—popularized by the so-called “Ralph Wiggum” plugin—demonstrated powerful self-healing coding loops.
The real prize, however, was not the Claude Code interface itself but the underlying Claude Opus 4.5 model. By spoofing the official client, third-party tools allowed developers to run large-scale autonomous workflows on Anthropic’s most capable reasoning model at a flat subscription price—effectively arbitraging consumer pricing against enterprise-grade usage.
As developer Ed Andersen noted on X, some of Claude Code’s popularity may have been driven by this very behavior.
For enterprise AI teams, the message is clear: pipelines built on unofficial wrappers or personal subscriptions carry significant risk. While flat-rate tools like OpenCode reduced costs, Anthropic’s enforcement highlights the instability and compliance issues they introduce.
Organizations now face a trade-off between predictable subscription fees and variable, per-token API costs—but with the benefit of guaranteed support and stability. From a security standpoint, the episode also exposes the dangers of “Shadow AI,” where engineers quietly bypass enterprise controls using spoofed credentials.
As Anthropic consolidates control over access to Claude’s models, the reliability of official APIs and sanctioned tools is becoming more important than short-term cost savings. In this new phase of the AI arms race, unrestricted access to top-tier reasoning models is no longer a given—it’s a privilege tightly guarded by their creators.