Today I came across this post in a Facebook group.
Now this man, he wasn't doing any spoofing or whatnot. He was actually doing a proper Anthropic API, using Claude to do coding.
He had spent over $10,000 worth of API costs over the course of two months
He wasn't doing anything wrong. He was just building.
Mid-session, his access was cut.
He re-subscribed and was still locked out. Then he went to sleep, and woke up to the same thing.
Anthropic never explained what happened.
This hits close to home because many years ago, Facebook suddenly banned my ad account for running ads that 'violated community guidelines'
...except, at the point of being banned, I haven't even ran any ads for 6 months straight.
Shrug.
The arbitrary-ness of large corporations, their bot review process, and the way they treat us is really no joke.
$10K is not a small amount at all, and in spite of that, Anthropic's auto-review process banned him.
This concerned me because I am building WITH Claude, on Claude Code.
And NLC is built on top of Claude Code.
I’ve been warned by my friends about the “dangers” of doing this.
But I’ve done my homework many times.
The answer is clear: All the accounts that were banned by Anthropic were basically people using Anthropic's OAuth to fuel a third-party app (like OpenClaw).
It is drastically different from NLC which is NOT, in fact, a ‘third party app’.
It’s a harness that comes with a very nice Web UI, which is a very common practice being done by Claude Code CLI users all around the world.
If you want to verify, just go look at GitHub - many projects with tens of thousands of stars have gone viral for doing the same thing.
The difference is basically this:
Spoofing: Steal the fuel (OAuth) to power a third party engine (3rd party app)
What is allowed: Using the fuel (OAuth) to power Anthropics OWN engine (Claude Code)
NLC IS Claude Code. We just pimp it up and make it look good + user friendly.
Think of NLC as the ‘chassis’ housing the engine that is Claude Code.
This knowledge, combined with many conversations plus research, made me feel assured.
But this post reminded me about something crucial:
We're building on land we don't own.
Anthropic has been known to do things we can’t explain.
Their models can be extremely smart one moment, complete idiots the next.
Their policies change at any given moment.
And they change their rate limits without so much as a warning.
One moment they will be nice to us, but in another moment they can sometimes be very… cold.
So instead of continue to build on land we don’t own, I’ve decided to prepare contingencies.
.
#1: A Permanent Backup
I'm installing a local AI model (Qwen 2.5) inside every NLC server.
This model runs locally, which means that NLC customers will all have permanent FREE AI to use in the form of Qwen 2.5 if they want to.
If Anthropic decides to cut us off, there will be backups running on every customer servers.
.
#2: Integrating OpenRouter
I've connected OpenRouter (a routing service) that users can plug in to switch to different AI provider if one goes down.
This is something that’s been known to be okay. Anthropic is not cool with people using their OAuth ‘fuel’ on 3rd party apps…
…but they’re more than happy to have others use other provider’s APIs to power Claude Code.
So by integrating OpenRouter, my users will have hundreds of models to choose from in the form of OpenRouter API.
.
#3: Making NLC Engine Agnostic
I'm beginning the work to remove our dependency on Claude Code.
Making it so that if I wanted to, I can unplug NLC from Claude Code and let it run on other agents.
Maybe Codex…or even OpenClaw (who knows).
I WOULD rather keep working with Claude Code, but Anthropic’s decisions sometimes really don’t make sense, so this has to be done.
All of that is in progress right now.
.
Today’s story, along with all the stuff I’m hearing online, reminds me how things can go wrong at an instant without any reasonable explanation.
We would like to think that companies would be fair to us since we’re their ‘customers’, but the reality is that large corporations aren’t actually obligated to fair to us.
The auto-review process kicked in, and that guy got banned. Simple as that.
No human was even involved.
This really triggers me to think about the future too, considering how dependent we ALL are on the big companies like Stripe, GHL, etc. for our businesses.
But that’s a story for another day.
What do you think about this situation?
Hit reply and let me know.
I read all emails.
FELIX