Skip to content

$ cat ~/blog/frontier-back-online.md

2026-07-01  · 2 min read ·  #ai #industry #notes

The frontier comes back online

Eighteen days after the government ordered Fable 5 suspended, it's back. What the outage looked like from the seat of someone who builds on these tools — and what it taught me about wiring a stack to a single model.

In June I wrote about the day the frontier got recalled — the US government ordered Anthropic to suspend Fable 5 and Mythos 5, its most capable models, over a national-security concern Anthropic publicly disputed. That post ended with a snapshot, not a verdict, because the facts were still moving.

Here’s the update: it’s back. I flipped my tools over to Fable 5 this week and it picked up mid-project like nothing happened. In fact, full disclosure — this post was drafted with it.

if you don’t follow AI closely

The short version, no jargon: a “frontier model” is the newest, most capable tier of AI — the one the labs just shipped. In June, the government used export-control authority (the same kind of rule that governs advanced computer chips) to order the top tier switched off while it examined a security claim. The company complied, disagreed on the merits, and kept its other models running. Now the top tier is available again.

If that sounds like it shouldn’t matter to a court reporting firm in St. Petersburg — it does. These tools are woven into how a lot of software gets built now, including mine.

what eighteen days offline looked like from here

Honestly? Less dramatic than the headlines. And that’s the interesting part.

The work didn’t stop. The rest of the lineup — Opus, Sonnet, Haiku — kept running, and the projects I ship kept shipping. This site went through a framework upgrade, a full audit, and a couple dozen merged pull requests during the gap, most of it built with the models that stayed up. The frontier tier is the sharpest tool in the drawer, but it turned out not to be the only one that cuts.

That’s not luck. It’s the same principle I build into everything: degrade gracefully. The photo wall on this site keeps a pixel-art fallback for when the live feed is down. The feed proxy keeps a last-good copy so an outage doesn’t blank the page. When I wrote in June that it’s “worth remembering the next time you wire your whole stack to a single model” — this is what I meant. The outage was the test. The stack passed.

the part I’ll keep

I still think the government should be able to stop a genuinely unsafe deployment, and I still think the bar Anthropic asked for — a process that’s transparent, fair, and grounded in technical facts — is the right one. I don’t have inside knowledge of how the dispute resolved, so I won’t pretend to. What I can report from my seat: the model went dark between one afternoon and the next, and it came back the same way. Both halves of that sentence should shape how you build.

So: welcome back to the sharpest tool. Good to have it. Still keeping the whole drawer.

— Andrew

Andrew Mayes — AI engineer & legal tech leader in St. Pete, FL. Writes between deploys. Supervised by Sushi, with Obiwan forever in memory. → more posts