Friday, June 12th, 2026
17:21
Anthropic's San Francisco Headquarters

The SecureDrop notification lit the screen in a room that was now barely lit. Most engineers were heading home. Some had already left the office. The message displayed on the screen, its title reading like a relic of the Cold War.

Subject: "Export Control Directive Emergency National Security Authorities".

The letter was a four-page document signed by US Secretary Lutnick.

Inside the open‑plan floor, a handful of ops engineers were still at their terminals monitoring the glow of the Fable 5 telemetry dashboards. The model had been live for exactly seventy‑two hours, a triumphant launch that had turned Anthropic’s usually quiet research lab into nothing less than a household name. API calls streamed across the screens in smooth, 200‑status green lines, handling hundreds of thousands of requests per second. Then the GC’s voice boomed out:

“Everyone stop. We have a federal export order. It’s not a drill.”

The directive was surgical and absolute: block all access to Fable 5 and Mythos 5 by any foreign national, whether sitting in Tokyo, London, or at a desk inside the United States — including Anthropic’s own foreign‑national employees. The company had no mechanism to dissect its user base by passport in real time. The only path to compliance was total shutdown… nothing less…

With a single terraform apply against the model‑router config, the routing layer that mapped the friendly name, claude‑fable‑5 to the massive inference clusters began to return… nothing…

Less than 30 minutes later, the first customer in Singapore could only see status: 404…

By 18:03, the status page was a sea of red, not green.

Amazon revoked IAM access, right in the middle of a deployment during their own Friday night change window.

Developers around the world who had been building agents on Fable 5’s new reasoning engine stared at error logs that simply read: model_not_found. The API that had been the most capable public AI in history evaporated faster than a cloud instance terminated by a budget alarm.

The reason for the order was left maddeningly vague: the government believed a jailbreak existed. Anthropic’s red‑teaming had pounded the safeguards for thousands of hours, with the US AI Safety Institute, the UK AISI, and tier‑one security firms all failing to find a universal break. Yet somewhere a rival had demonstrated a narrow, non‑universal technique to Commerce officials.

The letter didn’t elaborate. It didn’t have to. The full weight of export‑control law came down on a Friday night, and Fable 5 became the first large‑scale AI model ever ordered to die by its own government.

The paper was never even published. It sits now, locked inside a private repo. The entire industry, from bank CIOs to homelab hackers who had just containerized a local Fable‑5‑powered orchestrator… had learned a definitive lesson that evening. When the state sends a word, the frontier goes dark by dinner…

The struggle between the private company and the state has been ongoing since March 2026. There’s been a long uneasy history between the Trump administration and Anthropic. From a little over a mere $300 billion valuation to being declared a supply chain risk by the United States Pentagon to a company worth nearly one trillion dollars as of May 28th and the subsequent S-1 filing with the SEC for an IPO… Anthropic is, or maybe…was, a behemoth to be reckoned with. After this action, the gross revenue of the company has declined dramatically.

The company that called itself the ethical AI company that refused to allow the Pentagon to use Claude for “all lawful purposes” and was smeared as “woke radical left” by Donald Trump and his administration built a model so dangerous it required Glasswing access controls and was only able to be used by certain full time employees and some contractors at large financial institutions who worked in Cybersecurity. Anthropic used this “dangerous branding” to justify a nearly trillion-dollar company valuation. Little did they know that this very same branding would be weaponized against them by the very government they refused to fully serve. The investors who just poured nearly $100 billion into Anthropic are watching the flagship product get bricked less than two weeks after the funding round has closed.

You cannot write fiction this good. It’s as if the very universe itself is putting on an agorist performance.

Every single enterprise customer is getting a full crash course in “government can kill your AI provider”. OpenAI, Anthropic’s largest competitor, licks its chops, expediting its own S-1 filing. Dario Amodei is likely not getting much sleep this weekend.

This entire event is a lesson to all of us and should be heeded with care. National Security Theater, “copyright violations”, or any number of lesser reasons can now be used by the state to shut down any model. Anthropic caved instantly, without so much as a whimper or a blog post. Every company knows that any powerful government can simply flip the kill switch, all while its favored competitors proceed as if nothing bad is happening.

Agorism is about voluntary peaceful exchange. Agorism is about individual rights and expression. It’s about building parallel structures that do not require permission. Agorism is something that the ENTIRE AI space is severely lacking in. LunarWing intends to help change that.

As men and women of the dark forest, we already understand why self-hosted, locally run AI, and in fact, not just AI, but all software in general, is so important. There is no kill switch. There is no public cloud. There is no government boot to come down and destroy you by sending a simple letter.

Self-hostable. Free and open source. Secure. Resilient.

We are building LunarWing.

Let there be dark.

Everything here is licensed under AGPL-3.0, and the source lives on GitHub. Read along, file issues, send patches.

Talk soon. 🌒