Claude Fable 5 is suspended — what it means if you build on Claude

If you're calling Claude Fable 5 or Mythos 5 in any capacity — API, app, subscription — those calls are failing. They've been failing since June 12.

This is not a service outage. The US government issued an export control directive, and Anthropic shut both models down for every user worldwide that same evening. Nine days later, they're still offline with no confirmed restoration timeline.

Here's what happened and what to do.

What happened

Claude Fable 5 launched June 9, 2026. Three days later, at 5:21 PM ET on June 12, Anthropic received a directive from the US government and disabled Fable 5 and Mythos 5 immediately across all customers.

The stated reason: a jailbreak method — specifically, a technique involving asking the model to "read a specific codebase and fix any software flaws." Anthropic reviewed the demonstration and confirmed the technique exists, but found it narrow, non-universal, and already present in other models on the market.

Anthropic disagreed with the decision publicly:

"If this standard was applied across the industry, it would essentially halt all new model deployments for all frontier model providers."

They complied anyway. National security directive, no room to negotiate timelines.

What's still working

Every other Claude model is unaffected:

• Claude Opus 4.8 (claude-opus-4-8) — the current ceiling for reasoning and long-context work. Most Fable 5 workloads fit here.
• Claude Sonnet 4.6 (claude-sonnet-4-6) — faster and cheaper. The practical daily driver for most coding and writing tasks.
• Claude Haiku 4.5 (claude-haiku-4-5-20251001) — best for high-volume or latency-sensitive calls.

If your code targets claude-fable-5 or claude-mythos-5, swapping the model string to claude-opus-4-8 is the immediate fix. The API shape is unchanged.

If you're evaluating alternatives right now

OpenAI's GPT-5.5 Instant and Google's Gemini 3.5 Flash are both live and unaffected. If this is a moment to reconsider your model mix, ChatGPT vs Claude is the practical comparison, and which model to use by task walks through the decision without benchmarks you can't reproduce.

The broader point

This is the first time a government has forced a commercial AI model into suspension — not for a specific deployment, but for the model itself, everywhere, for everyone.

The lesson isn't "don't build on Claude." It's the same one that came up when Gemini CLI stopped serving individual users: the platform risk exists, and the teams that felt it least were the ones who had already thought about what breaks when a model disappears.

That means keeping model strings in config rather than scattered through code, having a swap-ready fallback in your model selection layer, and knowing which Anthropic tier you're actually on. None of that is heavy — it's the kind of seam that costs nothing to build in and saves real time when you actually need it.

Anthropic says they're working to restore access. No timeline has been confirmed.

---

For the day-to-day question of which model to reach for — Opus, Sonnet, Haiku, or something else — /learn/ai/choose-the-right-ai goes deeper.