The Model Disappeared

June 12. A letter from Commerce lands at Anthropic. Disable Fable 5 and Mythos 5 for all foreign nationals. Instead of segmenting access, Anthropic pulls both models offline for everyone. API calls start failing. Workflows stop.

Ten days later, both models are still dark.

I wrote about this today because the coverage keeps missing the point. The national security angle is interesting. The jailbreak that triggered it is interesting. Neither helps the team whose code review pipeline went from running to broken between lunch and end of day.

The companies that barely noticed had a fallback model already running. The ones still recovering went all-in on a single vendor’s capability and never tested what happens without it.

This is not new risk. It is vendor concentration, the same exposure that cost Delta $500 million during the CrowdStrike outage in 2024. What changed is that AI models now sit inside critical paths — overnight batch jobs, ticket routing, PR review — and governments have started pulling them offline by directive, not legislation. No warning period. No migration window.

The fix is boring. Abstraction layer. Tested fallback. Workflow logic separated from model identity. A few days of engineering against whatever June 12 cost the teams that skipped it.

The models will come back. The pattern won’t.