Claude Fable 5 Pulled by US Export Order — 72 Hours After Launch

Three days. Claude Fable 5 — Anthropic’s most capable model ever shipped to the public, posting 95% on SWE-bench Verified — was live for exactly 72 hours before the US government issued an export control directive on June 12 that forced Anthropic to pull it globally. For everyone. Including US users. Including Anthropic employees who hold foreign passports.

Here is what Fable 5 actually is, what the government directive says, what Anthropic says about it, and what developers building on Claude should do while this gets resolved.

What Claude Fable 5 Is

Anthropic launched Claude Fable 5 on June 9, 2026, alongside Claude Mythos 5, its restricted sibling for government-adjacent cybersecurity work. Fable 5 is the first publicly available model in Anthropic’s new “Mythos-class” tier — a category above the previous frontier that Claude Opus 4.8 (released May 28) occupied.

The benchmark gap is not close. Fable 5 posted 95.0% on SWE-bench Verified and 80.3% on SWE-bench Pro. The next best competitor on SWE-bench Pro is GPT-5.5, sitting at 58.6%. That is a 21.7-point gap — roughly twice the margin by which Claude Opus 4.8 led its generation. Across all eight coding benchmarks Anthropic published at launch, Fable 5 led with an average margin of 11.8 points:

Benchmark	Claude Fable 5	Next Best	Gap
SWE-bench Verified	95.0%	~74%	+21
SWE-bench Pro	80.3%	58.6% (GPT-5.5)	+21.7
FrontierCode Diamond	leads	baseline	+23.6
HLE (no tools)	leads	baseline	+13.7
Terminal-Bench	leads	baseline	+4.6

Beyond static benchmarks, Anthropic ran a long-horizon game-playing evaluation using Slay the Spire with persistent file-based memory. Fable 5 improved three times faster than Opus 4.8 as memory accumulated, and reached the final act three times as often. The large-context reasoning advantage — the same capability that powered the 8x engineering productivity multiplier at Anthropic — is structurally more pronounced in Fable 5 than in any previous public Claude release.

API Access and Pricing

The API model ID is claude-fable-5. At launch, access was available through:

• Direct API at api.anthropic.com
• Amazon Bedrock (natively supported at launch)
• Google Vertex AI
• Microsoft Azure Foundry

Pricing: $10 per million input tokens, $50 per million output tokens. Exactly double Claude Opus 4.8 on both sides. Context window: 1 million tokens by default, with up to 128k output tokens per request. The model supports vision, tool use, memory, compaction, and adaptive thinking.

For subscription users on Pro, Max, Team, and Enterprise plans, Fable 5 was available at no extra charge from June 9 through June 22, 2026, after which continued use would require usage credits at API rates. That 13-day promotional window was also, as it turned out, the entire window of public availability before the shutdown.

One hard constraint: Fable 5 and Mythos 5 carry 30-day data retention and are not available under zero data retention agreements. Enterprise customers with strict data-handling requirements who were planning to adopt Fable 5 need to factor that in alongside the compliance situation.

The Safety Architecture Anthropic Built

Getting Fable 5 to public availability required a new safeguard layer. Queries touching cybersecurity and biology — the two domains where Anthropic judged the capability uplift to be most consequential — are automatically rerouted to Claude Opus 4.8 rather than Fable 5. Anthropic says the safeguards trigger in fewer than 5% of user sessions on average.

Claude Mythos 5 is the model without those public-facing safeguards. It sits behind a vetted partner access gate and is the version made available to organizations doing government-adjacent security work. The Mythos Preview, which was gated behind Project Glasswing throughout April and May 2026, had already demonstrated capabilities that made even Anthropic pause: it autonomously identified and exploited a 17-year-old remote code execution vulnerability in FreeBSD with no human involvement after the initial request.

Fable 5 delivers a version of those capabilities with the safeguard layer applied. That distinction — same underlying model class, different deployment configuration — is central to understanding why the government’s shutdown directive covers both models.

The 72-Hour Timeline

The sequence is worth stating precisely, because the speed of what happened matters:

• June 9, 2026: Anthropic launches Claude Fable 5 and Mythos 5. Fable 5 goes live on all four major platforms and is available to all Claude API customers. Mythos 5 goes to vetted partners.
• June 12, 2026: The US government issues an export control directive. The directive states the government’s belief that a method exists to jailbreak Fable 5 and bypass its protective safeguards. No specific technical details about the alleged jailbreak are provided to Anthropic.
• June 12–13, 2026: Anthropic disables access to both Fable 5 and Mythos 5 globally. All API requests to claude-fable-5 and claude-mythos-5 return access-denied responses.

Anthropic has no real-time mechanism to verify the nationality of API requesters. Rather than attempt partial enforcement that the company could not reliably execute, it shut both models down entirely. Anthropic’s public statement described the directive as a “misunderstanding” and confirmed it is working to restore access. All other Claude models — Opus 4.8, Sonnet 4.6, Haiku 4.5, and prior versions — are not affected by the directive.

Reading the Government’s Position

The government’s concern, as stated, is that a foreign national with access to a jailbroken Fable 5 could extract capabilities that the safeguard layer was designed to block — primarily in cybersecurity and potentially in other sensitive domains. Whether such a jailbreak method actually exists is the factual crux of the dispute.

Anthropic’s position is that it does not. The company built the safeguard layer specifically to handle this class of risk and, by implication, believes the layer is not compromised. The government’s directive, as reported, did not provide technical evidence for the claimed jailbreak — which is why Anthropic is describing the situation as a misunderstanding rather than a confirmed vulnerability.

The export control framing is significant independent of whether the jailbreak claim is true. Export controls on AI models — as opposed to hardware like H100 GPUs — are relatively untested as a legal mechanism. The Great American AI Act, still working its way through Congress, includes provisions for exactly this type of frontier model access control. This directive, issued under existing executive authority rather than new legislation, may be testing the legal plausibility of that approach ahead of formal statute.

What Developers Should Do Right Now

If you were building against claude-fable-5, route to claude-opus-4-8 immediately. Opus 4.8 is unaffected, was the world’s top-ranked model on the Intelligence Index before Fable 5 launched, and runs at $5/M input and $25/M output — exactly half the cost of Fable 5. The capability gap is real but Opus 4.8 is not a fallback you are embarrassed to ship; it is still the best model most of the industry has ever had access to.

The more important architectural takeaway: any production pipeline with a single model ID as its only inference path has a fragility problem that this event just exposed. Use the AI Model Cost Calculator to model your three-tier fallback cost structure. The pattern that holds up under events like this has a primary model, a cost-equivalent fallback, and a high-volume budget tier — and the primary model should never be a model that has been in public production for fewer than 30 days.

For teams using Claude Code directly: the underlying Sonnet 4.6 model powering most Claude Code sessions is completely unaffected. Your development workflows continue unchanged. Use the AI Token Counter to audit what your agents are actually consuming per session before you route anything to Fable 5 on restoration, since the $50/M output price will hit differently at production scale than it did during the free window.

What’s Next

Anthropic has not given a restoration timeline. The most likely resolution path: Anthropic provides the government with technical evidence that no viable jailbreak of Fable 5’s safeguard layer exists, the directive is lifted, and access resumes. Based on the pace of similar export control clarifications in the hardware space, that is a process measured in days to weeks, not months.

The alternative — that a real jailbreak exists and the government’s concern is substantiated — would mean a longer pause while Anthropic patches the safeguard layer. That scenario also sets a precedent: the first documented case of a frontier model’s safety mechanism being publicly bypassed in a way that triggered government enforcement action. The implications for how frontier models are deployed going forward would be substantial.

The third possibility is the most bureaucratically plausible: the directive stands on technical grounds that have nothing to do with an actual jailbreak, gets resolved through legal and policy channels over several weeks, and access resumes with additional compliance attestation requirements baked in. The Great American AI Act process will absorb this as a test case either way.

Watch Anthropic’s news page for the official restoration announcement. The moment claude-fable-5 is back in the model registry, the 95% SWE-bench ceiling becomes your baseline again — not a benchmark you read about and then lost access to. Every product built on wowhow.cloud is designed to route to whatever Claude tier is currently available, pay once, ship forever.