The US Government Just Shut Down Anthropic's Most Powerful AI — What It Means for Every No-Code Builder

Anthropic launched the most capable AI model ever made publicly available on Monday. By Thursday evening, the US government had killed it — for everyone, everywhere, with no warning and no grandfather period.

Not deprecated. Not rate-limited. Disabled. Globally.

If you're a no-code builder whose app, workflow, or entire business runs on the Claude API, you woke up Friday morning to a hard lesson in single-provider dependency. And you're not alone. Bubble, Cursor, Bolt.new, Lovable, and hundreds of AI-native platforms rely on Anthropic's models somewhere in their stack. When the US Commerce Department pulled the trigger at 5:21pm ET on June 12, every single one of those dependency chains snapped simultaneously.

> TL;DR: Three days after launching Claude Fable 5 and Mythos 5 — models that posted 95% on SWE-bench Verified, beat GPT-5.5 by 21.7 points, and compressed months of engineering into days — the US government issued an unprecedented export control directive forcing Anthropic to disable both models globally. The shutdown affects paying enterprise customers, Amazon Bedrock users, and even Anthropic's own foreign-national employees. For no-code builders, this is the single-provider dependency crisis made real. The fix isn't finding another API key. It's building on platforms that abstract the model layer entirely.

What happened, exactly?

On June 9, Anthropic launched Claude Fable 5 — a Mythos-class model with safety classifiers, scoring 95% on SWE-bench Verified and beating GPT-5.5 by 21.7 points. Stripe described it as compressing "months of engineering into days" on a 50-million-line Ruby codebase. Alongside it came Claude Mythos 5, the same model without safety filters, restricted to Project Glasswing cyber defence partners.

On June 10, jailbreaker "Pliny the Liberator" posted a bypass on X — a multi-agent attack using Unicode homoglyphs and token-splitting to extract restricted outputs.

On June 12 at 5:21pm ET, Commerce Secretary Howard Lutnick sent a letter to Anthropic CEO Dario Amodei. The Bureau of Industry and Security had issued an export control directive — the first ever applied to a specific commercial AI model rather than hardware. It barred Anthropic from giving any foreign national access to Fable 5 or Mythos 5. That includes foreign nationals inside the US. That includes Anthropic's own foreign-national employees.

Anthropic can't do real-time nationality filtering on API calls. So it disabled both models for everyone. All queries now fall back to Claude Opus 4.8. Amazon Bedrock revoked access. Enterprise customers, government contractors, and researchers all cut off at the same moment.

Was the jailbreak actually dangerous?

Anthropic reviewed it and found it could identify "a small number of previously known, minor vulnerabilities" — the kind of thing other publicly available models do without any bypass. The company explicitly named OpenAI's GPT-5.5 as having the same capability.

Its public rebuttal was unusually blunt: "We disagree that software used by hundreds of millions of users should be blocked for this reason." It warned that pulling a commercial model over a narrow, non-universal jailbreak could "essentially halt all new model deployments for all frontier model providers."

This didn't come out of nowhere. In March, Defence Secretary Pete Hegseth labelled Anthropic a "supply chain risk" — the first time a US company has received that designation, historically reserved for firms based in adversarial countries. The Pentagon blacklisting followed Anthropic's refusal to allow Claude to be used for mass domestic surveillance and lethal autonomous weapons. A federal judge issued a temporary injunction on both, but the administration's posture was already clear.

The EU's response was sharp: Commission spokesman Thomas Regnier said the episode "further underlined Europe's need for technological sovereignty."

Why no-code builders should pay very close attention

If you build with no-code tools, here's the dependency chain.

Bubble announced AI features in April that depend on Claude for multi-turn editing and code generation. Cursor is built on a model-routing architecture where Claude handles the heavy lifting. Bolt.new and Lovable — platforms that have collectively raised hundreds of millions — use the Claude API for their core code-generation. Replit Agent uses it. StackBlitz uses it. The list goes on.

When the US government can kill your AI provider with a single letter sent at 5:21pm on a Friday, and that provider responds by disabling the model for every customer on Earth within hours, you have an operational risk no SLA can cover. You don't have a degraded service. You have a dead endpoint.

I've been writing about this all year. The Pentagon blacklisting in March was the warning shot. The OWASP top 10 for LLM applications lists model denial of service and supply chain vulnerabilities as genuine threat vectors. And we just covered the story about thousands of vibe-coded apps leaking corporate data because builders trusted default configurations they didn't understand.

The through-line is the same: the more you depend on a single model provider you don't control, the more fragile everything becomes.

What a survivable AI stack looks like

The builders who got through Friday without a scratch share one thing: they weren't betting on a single model.

Model diversity at the platform level. If your no-code platform only supports one AI model, that's a platform problem you inherit. The platforms worth betting on now abstract the model layer — where the provider can swap Claude for GPT-5.5, Gemini, or an open-weight alternative without you touching a single workflow.

Local and open-weight models as a fallback. AI founder Alex Finn called this a "wakeup call" to run local models on home GPUs. Most no-code builders aren't going to spin up a home server, but knowing that options like MiniMax M3 (open weights, frontier-class, runnable on your own hardware) exist changes the negotiating dynamic. Chinese labs are positioning their open models as the sovereignty play. Don't ignore that.

Multi-provider routing. Several platforms now offer model-routing layers that distribute requests across Anthropic, OpenAI, and Google depending on availability, cost, and capability. If you're building on a platform that doesn't offer this, ask why.

Structured no-code as the trust layer. This is the argument we've been making for months. Platforms built AI-core — where the AI is integrated into the architecture rather than bolted on through API calls — are inherently more resilient. When the model layer is abstracted away, model-switching is the platform's problem, not the builder's problem.

Stacker was designed this way from the start. The AI isn't a feature you toggle on — it's how the platform generates interfaces, manages data relationships, and handles authentication. If Anthropic goes dark, the platform routes to an alternative. The builder doesn't notice. That's the architectural decision that separates platforms that survive the next shutdown from the ones that won't. Glide and Webflow are making moves toward platform-level AI abstraction too, though neither is fully there yet. The platforms that get this right now won't need to send panicked emails the next time a government intervenes.

The takeaway

This won't be the last time a government pulls the plug on a frontier model. The Eastern Herald reports this directive "sets the operating template for a model-level export-control regime that is likely to be extended to OpenAI's GPT-6, Google DeepMind's Gemini 4 and the broader frontier-model peer group through the second half of 2026."

If you're a no-code builder, you have three things to do this week.

First, audit your dependency chain. Every tool you use, every API call your app makes — map where Claude sits in that stack. Second, ask your platform vendors what happens if their primary AI provider disappears overnight. If they don't have a clear answer, you have a clear problem. Third, start shifting toward platforms and tools that abstract the model layer. Not because Anthropic is going away — it isn't — but because the next shutdown might hit a different provider, and the one after that might not give anyone three days' warning.

The era of picking a favourite AI model and building your business on it is over. The era of platform-level model abstraction has just begun.