Anthropic's Best Model Got Banned in 72 Hours
Anthropic shipped the most powerful AI ever released to the public on June 9. By June 12 the US government made them pull it back from every foreign national on earth — including Anthropic's own engineers. The most capable model on the planet had a 72-hour public life. Oscar and I spent the episode figuring out who actually won that week, because it wasn't the people who got the model.
Here's how I read it, plus the stack we'd reach for.
Signal or Noise
Fable 5 and Mythos 5 ship at half the old frontier price — five days after Anthropic called for a global pause
Read that contradiction slowly. You don't cut prices in half on something you think should be paused. Signal, and a tell about how much the "pause" talk is really worth.
Export controls hit, and Anthropic disables access worldwide to comply
The whole point of a frontier model is that you can build on it. If a company can lose it overnight to a policy shift, that's a dependency you have to price in. Signal.
OpenAI files confidentially for its IPO at $730–850B — a week after Anthropic filed at $965B
Two unprofitable labs headed for public markets in the same window. The capital markets are about to have opinions about AI economics. Signal.
OpenAI wires its models and Codex into Oracle Universal Credits
Deleting the procurement step for every Oracle enterprise customer is a quiet distribution masterstroke. Signal — distribution beats benchmarks.
Stack Check
What I'd actually reach for right now:
- Claude Agent SDK — when you want to build the agent loop yourself and keep control of it.
- LangGraph + LangSmith — for graph-shaped agent workflows you need to trace and debug.
- Temporal — for the durable, long-running orchestration underneath, so a crashed step doesn't lose the whole job.
The lesson of the week: don't build your company on a model that can be switched off by someone who isn't you. Own the parts you can keep. Watch or listen below.
