Anthropic Clears Export Review, Restores Fable and Mythos and Ships Claude Sonnet 5

The bottleneck wasn't capability. It was paperwork. Anthropic's Fable and Mythos frontier models sat behind a regulatory hold — access restricted pending a federal export control review — while enterprise and API customers waited. That review has now concluded, and both models are back. Alongside the restoration, Anthropic has launched Claude Sonnet 5, slotting it into the mid-tier of a lineup that is, for the first time in months, fully intact.

What the Export Review Actually Meant

Federal export control reviews on AI models aren't routine nuisances — they signal that U.S. authorities are treating advanced frontier models as potentially dual-use assets subject to the same scrutiny as other controlled technologies. The hold on Fable and Mythos created a real gap: enterprise customers building on Anthropic's API couldn't access the full capability range the company had otherwise advertised.

The clearance now on record means U.S. authorities currently permit Anthropic's latest models for global commercial use, subject to existing AI export rules. That's a meaningful green light — not a blank check, but confirmation that Anthropic's frontier lineup clears the bar under today's regulatory framework. For competitors and policymakers watching how export controls apply to large language models, this resolution sets a data point.

Sonnet 5 and the Lineup It Completes

Claude Sonnet 5 enters with full commercial availability, positioned below Opus and above Haiku in Anthropic's tiered model hierarchy. That mid-tier slot matters operationally: it's typically where most production workloads land — capable enough for complex tasks, priced and structured for scale in a way that flagship models are not.

Anthropmic says Sonnet 5 brings improved tool use and safety guardrails compared with earlier Sonnet versions. Tool use improvements matter because agentic workflows — where models call external APIs, execute code, or chain tasks — are where enterprise adoption is accelerating fastest. Guardrail improvements reflect Anthropic's ongoing positioning around safety, a differentiator the company has leaned on since its founding.

With Fable, Mythos, and Sonnet 5 all live simultaneously, Anthropic now has its full frontier lineup back online for enterprise and API customers. The gap created by the export review is closed.

What Builders Should Take from This

Two things are worth parsing for anyone building on Anthropic's API.

First, regulatory exposure is now a real operational variable when choosing a model provider. The Fable and Mythos hold demonstrated that even a cleared, U.S.-based AI lab can face access interruptions tied to export review processes. Architects designing production systems should treat model availability as a risk factor — not just latency, cost, and context window.

Second, Sonnet 5's tool-use improvements are worth testing before assuming the capability profile mirrors earlier Sonnet releases. Mid-tier models in Anthropic's lineup often carry the most product-relevant performance gains because that tier is under the most competitive pressure — from OpenAI's GPT-4o class, from Google's Gemini mid-range, and from open-weight alternatives closing the gap. Anthropic had reason to make Sonnet 5 meaningfully better, not just marginally refreshed.

The Bigger Shift

This isn't just a product update or a regulatory footnote. The resolution of a federal export control review on commercial AI models — with clearance granted — marks an early precedent in what will become a recurring process as frontier capabilities advance. Authorities reviewed Anthropic's top-line models and approved their global commercial deployment under existing rules. That outcome won't hold indefinitely as model capabilities scale, and the next review cycle — for whatever comes after Opus — will be watched more closely.

For now, Anthropic's lineup is intact, Sonnet 5 is shipping, and the regulatory gate is open. Builders who paused evaluation of Fable and Mythos have no technical reason to wait.