Fable 5 is back: US lifts export controls after 18 days, Anthropic restores access globally July 1 — with a new industry jailbreak-severity framework

What resolved

Per Anthropic’s “Redeploying Fable 5” announcement (June 30) and reporting from Digital Trends, AIN, and TradingKey:

• Export controls lifted: the Commerce Department removed the restrictions on June 30, 2026.
• Return date: July 1, 2026, globally.
• Surfaces: Claude.ai, Claude Code, Claude Cowork, and the Claude Platform/API (cloud-provider re-enablement “forthcoming”).
• Usage terms: for Pro, Max, Team, and select Enterprise, Fable 5 is included for up to 50% of weekly usage limits through July 7; after that, usage-credit billing.
• The safeguard change: a new classifier targets the specific bypass technique Amazon reported — the trigger for the June 12 shutdown — and blocks it in over 99% of cases.
• The industry framework: Anthropic, Amazon, Microsoft, Google, and other Project Glasswing participants are proposing a shared jailbreak-severity scoring framework.

This bookends the timeline: June 12 shutdown → negotiations and the “trusted partners” framework → ID-verification infrastructure → Mythos 5 cleared for ~100 partners June 26 → full Fable 5 restoration July 1.

Why this matters

Three reads.

1. The most important output isn’t the reinstatement — it’s the severity framework. A model going back online is a return to the status quo. A shared, industry-wide method for scoring how dangerous a jailbreak actually is — across Anthropic, Amazon, Microsoft, and Google — is genuinely new infrastructure. Its four criteria (capability gain, breadth, ease of weaponization, discoverability) turn “is this jailbreak a national-security problem?” from a gut call into a rubric. That’s the kind of standard that could prevent the next 18-day scramble, or at least make government action more predictable. The standoff produced a governance tool the whole field can use.

2. It confirms the government-gated regime works in both directions. Over the past three weeks we’ve documented the US learning to restrict frontier models (the export-control shutdown, EO 14409, Mythos 5 to trusted partners). July 1 shows the reinstatement path: fix the specific vulnerability, demonstrate safeguards, propose a standard, get cleared. That round-trip — pull, patch, prove, restore — is now a template. For anyone building on frontier models, the lesson is that top-tier capability is contingent and negotiable, not permanent.

3. The weekly-quota model is the quiet business-model shift. Fable 5 comes back included in subscriptions (up to 50% of the weekly limit) rather than as a separate paid add-on — precisely what the leaked Claude Code strings foreshadowed. That’s a meaningful change to how frontier-tier access is priced: bundled with a cap, then metered. It makes the most capable model reachable for more subscribers, while protecting Anthropic’s inference economics. Expect this bundled-with-quota pattern to become standard for frontier tiers.

What it means for you

If you’re a Claude Pro/Max/Team subscriber: Fable 5 is back today, included for up to half your weekly usage through July 7 — a genuine window to use the strongest coding/reasoning model on real work before the usage-credit metering kicks in. The independent benchmarks that ran before the shutdown (#1 Intelligence Index, 95% SWE-bench Verified, 80.3% SWE-Bench Pro) are actionable again.

If you’re a Claude Code user: point it back at Fable 5 for the hardest long-horizon tasks — the capability that Cursor’s and Cognition’s CEOs praised at launch is available again, now with the bypass classifier hardened.

If you build on the API: plan for the July 7 transition to usage-credit billing, and note that cloud-provider (Bedrock/Vertex) re-enablement follows shortly rather than immediately.

For everyone: the practical default doesn’t change much — Opus 4.8 remained the workhorse throughout and still is for routine work. Fable 5 is the frontier option for when the extra capability earns its cost, and for vision-heavy work Gemini or GPT still lead.

The honest caveats

It’s a conditional return, not a clean victory. Fable 5 is back because a specific vulnerability was patched and safeguards were shown — not because the government concluded it overreached. The 50%-through-July-7 window and the usage-credit transition are real constraints. This is a negotiated reinstatement.

The severity framework is a proposal, not a standard yet. Anthropic, Amazon, Microsoft, and Google proposing a jailbreak-severity rubric is meaningful, but it isn’t adopted policy. Whether it becomes an actual industry or government standard is the thing to watch.

“99%+ blocked” is Anthropic’s figure. The new classifier’s effectiveness against the Amazon-reported bypass is Anthropic’s own measurement, pre-deployment. Real-world robustness will be tested by adversaries now that the model is public again.

The precedent cuts both ways. A frontier model can be pulled in hours and restored in weeks. That’s reassuring (resolution is possible) and unsettling (availability is contingent). Build accordingly.

What it changes for Pick Right readers

The headline is simple and good: the best coding/reasoning model of the cycle is available again, globally, from July 1 — with a generous included-usage window through July 7. We’ve updated the Claude review, the shutdown article, and the restoration tracker to reflect the resolution. The deeper takeaway is the one to carry forward: frontier-AI access is now governed — pullable, patchable, and restorable — and the industry just started building the rulebook (the severity framework) to manage it.

For the full arc, see the Fable 5 launch, the independent benchmarks, the government shutdown, the restoration negotiations, the Mythos 5 trusted-partners clearance, Executive Order 14409, the Cyber Jailbreak Severity (CJS) framework Anthropic detailed July 2 (the governance sequel to this restoration), and — the same-day companion story — OpenAI’s GPT-5.6 Sol/Terra/Luna preview, which is itself launching under the same government-gated regime.

Questions or corrections? Email Pick Right. Want the full list? See all news.