RIP Claude Fable 5.
Claude Fable 5 launched on Tuesday.
By Friday, it was gone.
A US government directive restricted foreign access to Anthropic’s newest models.
The result?
Anthropic disabled Fable 5 for every customer overnight.
People had upgraded to Claude Max specifically for Fable. Others were in the middle of projects when the model simply disappeared.
AI just had its “not your keys, not your coins” moment.
You do not own the model.
You do not own the weights.
You do not even own access.
You are renting permission from a company that can change the model, restrict your account, degrade its outputs, or be ordered to switch it off overnight.
Not your weights, not your model.
And no, the answer is not expecting everyone to buy a shiny new Mac Mini and run frontier models from their bedroom.
The answer is building alternative inference rails that do not depend entirely on one company remaining willing and legally able to serve you.
That is why I am so excited about projects like
@c0mputeAI .
The easiest explanation is BitTorrent for AI inference.
Instead of every request passing through one corporate gateway, independent workers share idle GPU capacity and run open models for users.
Developers connect through an OpenAI-compatible API.
If one worker disappears, another can take the job.
More importantly, the model itself does not disappear because one company loses permission to serve it.
Browser workers. Native GPU workers. Text and image models. Streaming. Tool calling. USDC payments.
c0mpute is building around the exact problem Fable exposed this week:
With centralized AI, you do not own the model.
You rent permission to use it.