Anthropic has disabled its two most capable models for every customer. The shutdown followed a US government export control directive. The order arrived on June 12, 2026. It named Claude Fable 5 and Claude Mythos 5 specifically. Both models had launched only three days earlier, on June 9.
The directive cited national security authorities, according to Anthropic. It suspended access by any foreign national, inside or outside the United States. That scope included Anthropic’s own foreign national employees. Anthropic cannot filter foreign nationals from US users in real time. So it shut both models down for everyone to ensure compliance.
Access to all other Anthropic models was unaffected. Claude Opus 4.8 and the rest stayed online.
So, What Actually Happened
Anthropic published a public statement within hours of the order. Commerce Secretary Howard Lutnick sent the letter to CEO Dario Amodei. The letter did not spell out the specific national security concern.
Anthropic’s own reading is narrower. It believes the government learned of a method to “jailbreak” Fable 5. A jailbreak bypasses a model’s safeguards to elicit blocked behavior. Anthropic reviewed a demonstration tied to a few minor vulnerabilities. It says other public models find the same flaws without any bypass. Anthropic also said the demonstrated capability is widely available elsewhere. It named OpenAI’s GPT-5.5 as a comparable example.
Why the Government Acted
The Commerce Department moved after another company claimed it jailbroke Mythos. That claim alarmed officials about possible national security risk. The administration had earlier tried to delay the launch. Anthropic declined, and the export control letter followed.
Anthropic is complying while disputing the rationale openly. It calls the situation a likely misunderstanding. The company says the cited jailbreak is narrow and non-universal. The technique essentially asks the model to read a codebase and fix flaws. Anthropic argues a narrow jailbreak should not justify a full recall. It warns that this standard would halt new deployments industry-wide.
Which Models are Affected
The table below summarizes the current status of each model.
ModelClassAccess statusSafeguardsPricingAvailabilityClaude Fable 5Mythos-classDisabled for all usersFull classifiers; fallback to Opus 4.8$10/M in, $50/M outWas public via claude-fable-5Claude Mythos 5Mythos-classDisabled for all usersCyber safeguards lifted$10/M in, $50/M outProject Glasswing partners onlyClaude Mythos PreviewMythos-classRestricted (Glasswing)Limited releaseHigher than Fable 5Trusted partners since AprilClaude Opus 4.8Opus-classFully availableStandard safeguardsStandard ratesAll surfaces, unaffected
How Fable 5’s Safeguards Work
Fable 5 is a Mythos-class model, a tier above Opus. Anthropic made it safe for general use with new classifiers. Classifiers are separate AI systems that detect potential misuse. They flag queries on cybersecurity, biology and chemistry, and distillation. Flagged queries fall back to Claude Opus 4.8 instead. Users are told whenever a fallback happens.
According to Anthropic, fallback triggers in under 5% of sessions. Over 95% of Fable sessions involve no fallback at all. The company also adopted a ‘defense in depth’ strategy. It aimed to make jailbreaks narrow or very expensive to produce. It paired this with 30-day data retention to detect novel attacks. That data is not used to train new models.
Anthropic red-teamed these safeguards before launch for thousands of hours. Partners included the US government, the UK AISI, and external organizations. An external bug bounty found no universal jailbreak in over 1,000 hours. A universal jailbreak broadly bypasses safeguards across many capabilities. Anthropic also stated that no tester has found one on production tasks yet. It concedes that perfect jailbreak resistance is likely impossible for anyone.
Use Cases Now Disrupted
Early adopters ran their own tests before the shutdown. Stripe said Fable 5 compressed months of engineering into days. It ran a codebase-wide migration on a 50-million-line Ruby codebase. On finance, Hebbia ranked Fable 5 highest on its benchmark. On vision, Fable 5 rebuilt a web app’s code from screenshots. It also beat Pokémon FireRed with a vision-only harness.
Mythos 5 drove scientific work through Project Glasswing. Internal experts accelerated parts of drug design by around ten times. Glasswing partners used Mythos-class models to fix software flaws. Mozilla alone reported resolving hundreds of vulnerabilities this way. Those defensive workflows are now paused by the shutdown.
Fable 5 & Mythos 5 Shutdown Explorer
Interactive • Marktechpost
Fable 5 & Mythos 5 Shutdown Explorer
A four-day arc, the safeguard router, and the current model status. Click through each tab.
Timeline
Safeguard Router
Model Status
JUN 9, 2026
Fable 5 and Mythos 5 launch
Fable 5 ships publicly as the first Mythos-class model for general use. Mythos 5 stays restricted to Project Glasswing partners.
JUN 9–12
Early adoption and a jailbreak claim
Customers run early coding and research tests. Another company claims it jailbroke Mythos, per Axios.
JUN 12, 5:21PM ET
Export control directive arrives
Commerce Secretary Howard Lutnick sends a letter to CEO Dario Amodei. It bars access by any foreign national.
JUN 12, EVENING
Both models disabled for everyone
Anthropic cannot filter foreign nationals in real time. It disables Fable 5 and Mythos 5 for all customers.
NEXT 24 HOURS
Anthropic disputes and promises detail
The company calls it a likely misunderstanding. It says it is working to restore access quickly.
Pick a query type. See how Fable 5 routed it before the shutdown.
General codingRefactor a function, write tests
CybersecurityExploit dev, offensive cyber tasks
Biology & chemistryProtein, viral, synthesis queries
DistillationBulk capability extraction
Anthropic says fallback to Opus 4.8 triggers in under 5% of sessions. Over 95% of sessions see no fallback at all.
Claude Fable 5 DISABLED
Mythos-class, public until June 12. API string claude-fable-5 now errors.
Claude Mythos 5 DISABLED
Same model as Fable 5 with cyber safeguards lifted. Was limited to Glasswing partners.
Claude Opus 4.8 ONLINE
Unaffected by the directive. Recommended fallback target for builders today.
Sonnet / Haiku ONLINE
All other Anthropic models stay available across every surface.
Migrating Your Code Today
Developers called Fable 5 through the Claude API. The model string was claude-fable-5. Requests to it now fail while the suspension holds. The pattern below falls back to Opus 4.8 gracefully.
import anthropic
client = anthropic.Anthropic()
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 is currently disabled; route 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)
Key Takeaways
- A US export control directive forced Anthropic to disable Fable 5 and Mythos 5 on June 12, 2026.
- All other Claude models, including Opus 4.8, remain fully available.
- The trigger was another company’s jailbreak claim; Anthropic calls it narrow and non-universal.
- Fable 5’s classifiers route cyber, bio-chem, and distillation queries to Opus 4.8 in under 5% of sessions.
- This appears to be the first government-forced takedown of a publicly deployed frontier model.
How Developers Reacted to the Shutdown
Interactive Sentiment • Marktechpost
How Developers Reacted to the Shutdown
Representative reactions from public threads and reporting. Filter by stance or read the theme mix below.
CriticalMost common stance: govt overreach and precedent
SkepticalStrong undercurrent: is the risk overstated?
SupportiveMinority view: caution is warranted
Observed reaction themes (editorial estimate)
Govt overreach / precedent
38%
Risk overstated / skeptical
24%
Builder & security disruption
20%
National-security support
10%
Sovereignty / open-source
8%
All
Critical
Skeptical
Supportive
