← ALL UPDATES
CLAUDE NEW June 12, 2026

Claude Fable 5 Launched and Suspended in 3 Days — What Happened

WHAT CHANGED

Claude Fable 5 launched June 9 as the first publicly available Mythos-class model. On June 12 at 5:21 PM ET, the US government issued an export-control directive ordering Anthropic to suspend access for any foreign national worldwide. Because Anthropic cannot filter by nationality in real time, Fable 5 + Mythos 5 were disabled for ALL customers globally. As of June 16, both remain unavailable. Other Claude models (Opus 4.8, Sonnet 4.6, Haiku 4.5) are unaffected.

WHY IT MATTERS

This is the first-ever US government recall of a frontier AI model. The directive cited alleged jailbreak vulnerability. Anthropic disputes the severity — says the demonstrated technique only found 'small number of previously known, minor vulnerabilities.' On June 15, Anthropic senior leaders met Trump admin officials in DC. No resolution reached. For builders: every production system that depended on Fable 5 broke June 13. The new model-availability risk is real.

HOW TO USE IT

The claude-fable-5 API string returns errors. Recommended fallback: claude-opus-4-8 for long-horizon coding, claude-sonnet-4-6 for routine production traffic, claude-haiku-4-5 for classification/routing. Action items: log model_id on every call, add non-Anthropic backup provider, freeze new launches that hard-depend on single frontier model for next 2 weeks.

CLAUDE / TYPESCRIPT 
// BEFORE June 12 — broken now
// model: "claude-fable-5"

// CURRENT FALLBACK (June 16, 2026)
const response = await anthropic.messages.create({
  model: "claude-opus-4-8",  // safe, available
  max_tokens: 8192,
  messages: [{ role: "user", content: prompt }]
});
claudefable-5mythos-5suspensionanthropicexport-controlbreaking
ORIGINAL SOURCE
https://www.anthropic.com/news
VIEW ORIGINAL SOURCE →

The 3-day launch and suspension reframes how to think about frontier AI dependency. Fable 5’s benchmarks at launch were genuinely strong (80.3% SWE-Bench Pro vs Opus 4.8’s 69.2%, GPT-5.5’s 58.6%). Stripe completed a 50M-line migration in a day. Hex hit 90% on its core analytics benchmark — first model ever.

All of that capability is now offline indefinitely. The takeaway for builders is structural: a single model dependency is now a single point of failure not just for technical or pricing reasons, but for regulatory ones too.

What’s still working: → Claude Opus 4.8 — best fallback for long-horizon coding work → Claude Sonnet 4.6 — 1.7x cheaper, handles most production traffic → Claude Haiku 4.5 — classification and routing layer → Claude Code, Design, Cowork — all operational on Opus 4.8 backend

Per Totalum’s incident response guide: “the new product launches that hard-depend on a single frontier model” should freeze for the next 2 weeks until the dust settles. Add a non-Anthropic backup provider. Log model_id on every call so you can rebuild your fallback decision logic.

← BACK TO UPDATES