Your $200/month Claude Pro subscription just became worthless. You can’t use it with OpenCode anymore. You can’t use it with Cursor. You can’t use it with anything except Anthropic’s official clients.
As of 7 hours ago, Anthropic updated their legal and compliance page with language that effectively bans all third-party tools from using OAuth tokens obtained through Pro, Max, or Free accounts.
The relevant section is buried under “Authentication and credential use”:
OAuth authentication (used with Free, Pro, and Max plans) is intended exclusively for Claude Code and Claude.ai. Using OAuth tokens obtained through Claude Free, Pro, or Max accounts in any other product, tool, or service — including the Agent SDK — is not permitted.
Let me translate this from corporate legalese into plain English:
If you’re using OpenCode, Cursor, or any third-party client with your Pro account credentials, you’re now violating Anthropic’s Terms of Service. They reserve the right to ban you without notice.
The Pattern: Enshittification in Real-Time
This isn’t surprising. It’s textbook enshittification — a term coined by Cory Doctorow to describe how platforms gradually degrade user experience to extract more value.
Phase 1: Launch with generous terms to attract users and developers.
Phase 2: Build dependency — get people to integrate your tool into their workflows.
Phase 3: Lock down. Raise prices. Restrict access. Extract rent.
We’ve seen this play out with:
- Twitter API: Went from free to $42,000/month overnight
- Reddit API: Killed third-party clients to force official app usage
- Spotify API: Gradually restricted features until it was useless for indie devs
Now it’s Anthropic’s turn. And if you think OpenAI won’t do the same the moment they feel secure enough, you’re not paying attention.
The Real Reason: Price Discrimination, Not Security
Anthropic will tell you this is about “security” or “preventing abuse.” Don’t buy it.
This is about price discrimination. They’ve created two tiers:
| Tier | Price | Allowed Clients | Target |
|---|---|---|---|
| Consumer (Free/Pro/Max) | $0-$200/mo | Official clients ONLY | Individual users |
| API (Pay-per-token) | ~$15/million input tokens | Any client | Enterprises & builders |
The math is brutal:
- Pro Plan: $200/month for ~1M tokens (effective rate: ~$15/million tokens)
- API Plan: $15/million input tokens, but you need to build your own client
By forcing Pro users into official clients, Anthropic achieves two things:
- Brand control: Every interaction happens inside “Claude Code” or “Claude.ai” — reinforcing brand association
- Upsell pressure: Enterprises that need flexibility must migrate to API pricing (which, conveniently, is much more expensive at scale)
One commenter on Hacker News put it bluntly:
They know consumers are price sensitive, and that companies wanting to use their APIs to build products so they can slap AI on their portfolio and get access to AI-related investor money can be milked.
The Backlash Is Already Here
The Hacker News thread has 414 comments in 7 hours — one of the most active discussions this year. The sentiment is overwhelmingly negative:
I don’t see how it’s fair. If I’m paying for usage, and I’m using it, why should Anthropic have a say on which client I use?
I pay them $100 a month and now for some reason I can’t use OpenCode? Fuck that.
I’m only waiting for OpenAI to provide an equivalent ~100 USD subscription to entirely ditch Claude.
Everyone else (Codex CLI, Copilot CLI etc…) is going opensource, they are going closed.
This hostile behaviour is just the last drop.
Opus 4.6 genuinely seems worse than 4.5 was in Q4 2025 for me.
This is the first time I’ve really felt it with a new model to the point where I still reach for the old one.
That last quote is critical. Anthropic isn’t just restricting access — users are reporting model quality degradation with Opus 4.6 compared to 4.5. When you combine declining quality with hostile restrictions, you get churn.
What This Means for the AI Ecosystem
1. The MCP Movement Will Accelerate
One commenter predicted:
I’m predicting that there would be a new movement to make everything an MCP. It’s now easier to consume an API by non-technical people.
The Model Context Protocol (MCP) was designed exactly for this scenario — standardized, vendor-neutral tool integration. Expect a surge in MCP server adoption as developers look for escape hatches from vendor lock-in.
2. Open Source Clients Will Pivot to API-Only
Projects like OpenCode (Anthropic’s own open-source CLI, ironically) and Cursor will need to either:
- Require users to bring their own API keys (pay-per-token)
- Build wrapper services that absorb the cost (unsustainable)
- Abandon Claude integration entirely
My bet: most will add OpenAI/Gemini as primary options, with Claude as a secondary “bring your own API key” feature.
3. The “Open” in AI Was Always an Illusion
Let’s be honest: there is no open AI ecosystem. There never was.
- Models: Closed weights, closed training data, closed evals
- APIs: Rate-limited, terms-restricted, revocable at any time
- Clients: Now even OAuth tokens are locked to official apps
The only truly open layers are:
- Inference frameworks (vLLM, TGI, llama.cpp)
- Open-weight models (Llama, Mistral, Qwen) — but these lag frontier capabilities
- Protocols (MCP, A2A) — still emerging
If you’re building a business on top of proprietary AI APIs, you’re building on rented land. The landlord can change the locks whenever they want.
What Should You Do?
For Individual Developers
Diversify your model access: Don’t rely on a single provider. Set up API keys for OpenAI, Anthropic, and Google. Use a router like LiteLLM to abstract differences.
Consider self-hosting: For non-frontier use cases, models like Qwen 2.5 72B or Llama 3.1 405B (if you have the hardware) give you full control.
Vote with your wallet: If you’re a Pro user who relies on third-party clients, cancel and switch to API billing. Yes, it’s more expensive. But it’s the only way to maintain flexibility.
For Companies Building AI Products
Never build on OAuth: Always use API keys with proper billing. The slight cost premium is insurance against policy changes.
Abstract your model layer: Use a provider-agnostic framework (LangChain, LlamaIndex, or custom) so you can swap providers without rewriting your entire codebase.
Keep an open-source escape hatch: For critical features, maintain compatibility with open-weight models. You may need to fall back to them if API access is restricted or priced out.
For the Community
Support open protocols: Contribute to MCP, A2A, and other vendor-neutral standards. These are the only defense against lock-in.
Fund open models: Donate to organizations like EleutherAI, Stability AI, or Mistral. The more competitive open models become, the less leverage proprietary providers have.
Call out enshittification: When platforms degrade terms, amplify the backlash. Silence enables abuse.
The Irony Nobody’s Talking About
Here’s the kicker: Anthropic was founded on the principle of “AI safety” — the idea that AI development should be constrained to prevent harm.
But what we’re seeing isn’t safety. It’s rent-seeking.
Real safety research would mean:
- Open evaluation benchmarks so the community can verify claims
- Transparent incident reporting when models cause harm
- Collaboration on governance frameworks
Instead, we get:
- Closed evals
- NDAs for enterprise customers
- Legal restrictions on how users can access the tool they paid for
If Anthropic truly believed in AI safety, they’d welcome third-party auditing and tooling. Restricting access to official clients doesn’t make AI safer — it makes Anthropic more money.
The Bottom Line
Anthropic’s OAuth lockdown is a watershed moment. It’s the clearest signal yet that the “wild west” era of AI development is over. The frontier model providers have decided: they will control the stack, end to end.
This isn’t inherently evil. Apple controls iOS end-to-end, and it works (for Apple). But it means the dream of an open, composable AI ecosystem — where anyone can build tools on top of frontier models — was always a fantasy.
The question now is: what comes next?
Will the community rally around open-weight models and protocols? Will OpenAI resist the temptation to lock down? Will a new player emerge with genuinely open terms?
Or will we accept a future where AI development is controlled by a handful of companies who can change the rules whenever they want?
I know which future I’m betting on. And it doesn’t involve trusting Anthropic’s Terms of Service.
Your move, builders.
What’s your take? Are you sticking with Claude despite the restrictions, or is this the push you needed to explore alternatives? Drop a comment — I read every single one.
And if you found this useful, share it with someone who’s still under the illusion that “open AI” exists. They need to wake up.