Claude Fable 5 and Mythos 5 Suspended by US Government: The Full Story

Three days after launching its most powerful model ever, Anthropic was forced by US government export control to shut down Claude Fable 5 and Mythos 5 for every customer. Here is what happened and why.

On June 9, 2026, Anthropic launched Claude Fable 5 — the most powerful AI model ever made available to the general public. It scored 95% on SWE-bench Verified, could one-shot full software applications, and was described by partners as a leap beyond anything previously available.

Three days later, on June 12 at 5:21 PM ET, it was gone. Not deprecated. Not retired with a migration window. Gone for every customer on the planet.

The US government issued an export control directive ordering Anthropic to suspend all access to Fable 5 and its restricted sibling Mythos 5 for any foreign national. Because Anthropic could not reliably distinguish foreign nationals from domestic users in real time, the practical result was a global shutdown of both models for all customers.

This is the first time in history that a US government export control order has targeted a specific commercial AI models deployed inference — not chips, not datasets, not source code. The model itself.

The Timeline: Launch to Shutdown in 72 Hours

• June 9, 2026 (10 AM PDT): Anthropic launches Claude Fable 5 as generally available. First Mythos-class model open to the public. Pricing: $10/$50 per million input/output tokens — double Opus 4.8.
• June 12, 5:21 PM ET: Anthropic receives the export control directive citing national security authorities.
• June 12, ~11 PM UTC: Global shutdown. All API calls to claude-fable-5 and claude-mythos-5 begin returning errors.
• June 13, 2026: Anthropic publishes public statement. No timeline for restoration given.

Why Did the US Government Ban Fable 5?

The government cited awareness of a method of bypassing, or jailbreaking Fable 5. The concern: a technique allowing the model to read a specific codebase and identify software vulnerabilities — using Fable 5s world-class code analysis for offensive cybersecurity.

Anthropic disputes this. In their official statement:

We reviewed a demonstration of this specific technique being used to identify a small number of previously known, minor vulnerabilities. These all appear relatively simple, and we have found that other publicly-available models — specifically GPT-5.5 — can discover them without requiring a bypass.

What Mythos 5 Was

Fable 5 and Mythos 5 are the same underlying model packaged two ways. Fable 5 was publicly available with safety classifiers that fallback to Opus 4.8 for cybersecurity, biology, chemistry queries. Mythos 5 was the same model with safeguards lifted, available only to Project Glasswing partners. Through Glasswing, 150 organizations across 15 countries had used Mythos Preview to find over 10,000 critical vulnerabilities.

The Jailbreak Debate

• Anthropic: Perfect jailbreak resistance does not exist. Non-universal jailbreaks apply to all frontier models. This standard would halt all new model deployments.
• Government: Fable 5s vulnerability-at-scale capability crosses the threshold for export control, regardless of competitor parity.

What This Means for Builders

Fallback options: Opus 4.8 (88.6% SWE-bench, $5/$25) is the closest replacement. Sonnet 4.6 (79.6%, $3/$15) for latency-sensitive work. Builders who hard-coded claude-fable-5 with no fallback had production outages within hours.

Three Key Implications

1. Government intervention in AI availability is now a real operational risk. Generally available can be revoked within hours.
2. Model redundancy is now a security posture, not just cost optimization.
3. You cannot distinguish an export control shutdown from a normal API outage without reading provider status pages.

Fable 5 Benchmark Context

• Fable 5 (SUSPENDED): 95.0% SWE-bench Verified, 80.3% Pro. $10/$50 per MTok.
• Opus 4.8: 88.6% SWE-bench Verified, 69.2% Pro. $5/$25 per MTok.
• Sonnet 4.6: 79.6% SWE-bench Verified. $3/$15 per MTok.
• Haiku 4.5: 73.3% SWE-bench Verified. $1/$5 per MTok.

The Fable 5 to Opus 4.8 gap was the largest single-generation leap in Claude history — 6.4 points on Verified, 11.1 on Pro.

What Happens Next

Anthropic is complying while disputing the rationale. No restoration timeline given. The June 22 billing cliff (Fable 5 moving to credits-only) is moot while suspended. Monitor Anthropics status page for updates.

Sources: Anthropic official statement, TechCrunch, CNBC, Bloomberg, ChatForest Builder Guide, TokenMix Research Lab, MorphLLM benchmarks. All benchmarks verified from primary sources. Published June 13, 2026.