Anthropic disables its most capable two models for all customers. A US export control directive prompted the shutdown. Order arrived June 12, 2026. Claude Fable and Claude Mythos were specifically mentioned. The models were launched just three days before, on 9 June.
Anthropic reports that the directive was backed by national security authorities. This directive prohibited access to any foreign nationals, in the United States or abroad. Anthropic’s foreign employees were included in the scope. Anthropic is unable to filter US users from foreigners in real time. To ensure compliance, it closed both models.
All other Anthropic models remained online. Claude Opus 4.8 remained available.
Then, what actually happened?
Anthropic released a statement to the public within hours after receiving an order. Dario Amedei received the letter from Commerce Secretary Howard Lutnick. The letter didn’t specify the national security concerns.
Anthropic is more limited in its own interpretation. It thinks the government discovered a new method of “jailbreak” Fable 5 An jailbreak is a way to bypass a model’s safety measures and elicit restricted behavior. Anthropic looked at a video demonstration that had a couple of minor flaws. Other public models have the same problems, it says. Anthropic said that the capability demonstrated is available in many other places. OpenAI GPT 5.5 is cited by Anthropic as a similar example.
Why Government Acted
A company claiming to have cracked Mythos prompted the Commerce Department’s response. The officials were alarmed by this claim, which could pose a national security threat. The administration attempted to delay launch earlier. The letter was sent to Anthropic after the refusal.
The rationale is openly contested by Anthropic, but it complies. The company calls it a possible misunderstanding. According to the company, jailbreaks are not universal and only applicable in certain situations. It is a technique that asks a model to read code and correct flaws. Anthropic claims that a small jailbreak does not warrant a complete recall. The company warns this standard could stop new deployments across the entire industry.
What Models Are Affected?
Below is a table that summarizes each model’s status.
| Model | The Class | Access status | Safety Measures | Pricing | Accessibility |
|---|---|---|---|---|---|
| Claude Fable 5 | Mythos-class | All users are disabled | If you want to use Opus 4.8, then opt for the full-classifiers. | Spend $10/M and get $50/M | The public can be reached via claude-fable-5 |
| Claude Mythos 5 | Mythos-class | All users are disabled | Cyber safeguards lifted | 10/M$ in, 50/M$ out | Project Glasswing Partners Only |
| Claude Mythos Preview | Mythos-class | Restricted Glasswing | Limit release | The Fables 5: A Higher Standard | Trusted partner since April |
| Claude Opus 4.8 | Opus-class | Unavailable | Standards of protection | Rates for Standard Services | All surfaces, unaffected |
Fable 5 Safeguards: How they Work
Fable 5, a model of the Mythos Class, is an echelon above Opus. Anthropic improved its general usability with the new classifiers. They are AI systems which detect possible misuse. The flagging includes queries related to cybersecurity, biology, chemistry and distillation. Flagged queries are redirected to Claude Opus 4.8. When a user is informed of a possible fallback, they are alerted.
Anthropic claims that fallback is triggered in less than 5% of all sessions. Fable’s sessions are 95% fallback-free. The company also adopted a ‘defense in depth’ strategy. In order to reduce the cost of jailbreaks, it aimed at making them narrower or more difficult to perform. This was paired with 30-day retention of data to detect new attacks. This data will not be used for training new models.
Before launch, Anthropic spent thousands of working hours red-teaming these safeguards. The US government, UK AISI as well as external organizations were partners. A bug bounty conducted by an external organization found that no universal jailbreak was detected in more than 1,000 hours. The universal jailbreak bypasses many safeguards. Anthropic has also said that they have not yet found any testers who can find a jailbreak in production. Anthropic admits it is unlikely that anyone will ever be able achieve perfect resistance to jailbreak.
Use Cases now Disrupted
Early adopters conducted their own testing before shutdown. Stripe claimed that Fable 5 consolidated months of development into a few days. It performed a migration of the entire Ruby codebase, which consists of 50,000,000 lines. Hebbia’s benchmark ranked Fable 5, the best in finance. Fable 5 rebuilt the code of a web application from screenshots. It also beat Pokémon FireRed with a vision-only harness.
Mythos 5 fueled scientific research through Project Glasswing. Experts within the company accelerated drug development by ten to twenty times. Glasswing’s partners have used Mythos models to correct software bugs. Mozilla reported that it had fixed hundreds of software flaws this way. These defensive workflows have been paused due to the shutdown.
Fable 5 & Mythos 5 Shutdown Explorer
Interactive • Marktechpost
Fable 5 & Mythos 5 Shutdown Explorer
The current status of the model, a four-day forecast, the safety router and the current state. You can click through the tabs.
Timeline
Secure Router
Model Status
The JUN 9th, 2026
Fable 5 launches alongside Mythos 5.
Fable 5, the first Mythos class model, is now available for public use. Mythos 5 is only available to Project Glasswing Partners.
JUN 9–12
The early adoption claim and the jailbreak claim
Customers conduct early testing and coding. Axios claims that Mythos was jailbroken by another company.
JUN 12, 5 :21 PM ET
Export control directive arrives
Howard Lutnick, Secretary of Commerce sends Dario Amodei a formal letter. This letter bars foreigners’ access.
JUN 12, EVEING
All models of the two-model wheelchair are accessible to everyone
Anthropic does not have the ability to screen foreign nationals live. Fable 5 and Mythos 5 are turned off for all users.
NEAREST 24 HOUR
Details of Anthropic disputes, promises and details
This is what the company says. The company claims to be working hard to quickly restore access.
Pick a query type. Fable 5’s routed the query before it was shut down.
General CodingWrite tests, refactor the function
CybersecurityExploit dev, offensive cyber tasks
Biology & chemistryProtein, viral, synthesis queries
DistillationBulk capacity extraction
In less than 5% of all sessions, Anthropic reports that Opus 4.8 is used as a fallback. More than 95% sessions have no fallback.
Claude Fable 5 DISEABLED
Mythos class is open to the public from June 12 until then. The API string claude-fable-5 There are errors.
Claude Mythos 5 DISEABLED
Cyber safeguards removed. Same as Fable 5. Glasswing only.
Claude Opus 4.8 ONLINE
Directive not applicable. Build to the target recommended for today.
Sonnet / Haiku ONLINE
The other Anthropic designs are available on all surfaces.
Upgrade Your Code Now
Fable 5 was called by developers through the Claude API. Model string was claude-fable-5. The suspension is still in place and requests to the system will fail. This pattern gracefully falls back to Opus 4.
Anthropocentric import
Anthropic = client()
Def ask(prompt (primary="claude-fable-5", fallback="claude-opus-4-8"):
for model in (primary, fallback):
try:
return client.messages.create(
model=model,
max_tokens=1024,
messages=[{"role": "user", "content": prompt}],
)
except anthropic.APIStatusError as err:
# Fable 5, currently unavailable; please go to Opus 4.8.
print(f"{model} unavailable: {err.status_code}")
raise RuntimeError("No model available")
print(ask("Summarize this change log.").content[0].text)
The Key Takeaways
- Anthropic was forced to deactivate Fable 5 and Mythos 5 by June 12th 2026 due to a US export control directive.
- Opus 4.8 is still fully available, as are all the other Claude Models.
- Anthropic claims that the trigger was a claim made by another company about jailbreak.
- Fable 5’s classifiers send cyber, biochem and distillation questions to Opus in less than 5% of session.
- It appears that this is the first time a government has forced a model to be taken down.
What Developers Have to Say About the Shutdown
Interactive Sentiment • Marktechpost
What Developers Have to Say About the Shutdown
Reactions from the public and media. Sort by viewpoint or view the mix of themes.
The Critical LanguageOverreach by the government and precedent are two of the most common positions.
SkepticalIs the threat overstated or not?
SupportiveWarning: Minority View
Editorial assessment of observed reaction themes
You can also read about the advantages of using
The Critical Language
Skeptical
Supportive
Sources

