The US government has ordered Anthropic to suspend access to its latest Fable 5 and Mythos 5 models for all users, including foreign nationals, citing national security concerns. While existing Claude models remain unaffected, the cutting-edge models designed for long-running, high-performance agent tasks are now unavailable, causing significant disruption for developers and enterprises relying on them.
📑Table of Contents
US Government Export Control Directive: Content and Background
On June 12, 2026, the US government instructed Anthropic to immediately suspend access to Fable 5 and Mythos 5 for all users. The directive is based on national security grounds, with particular concerns over potential misuse in cybersecurity and bio-chemical domains. The order explicitly includes foreign users and extends to API-based access.
Fable 5 represents Anthropic’s Mythos-class model optimized for long-horizon and complex tasks, significantly outperforming previous Opus 4.8 in agentic workflows. The government focused on the advanced agent capabilities of these models when placing them under export controls. Anthropic stated in its official announcement that even the restricted Mythos 5, previously available only to trusted partners via Project Glasswing, has been impacted.
Impact on Fable 5 / Mythos 5 and Differences from Existing Claude Models
Fable 5, previously available via public API, is now inaccessible to both new and existing users. Mythos 5, the safeguards-removed variant for Glasswing partners focused on cyber use cases, faces the same restrictions. Legacy models such as Claude Opus 4.8 and Sonnet series are not affected and remain fully operational.
In benchmarks, Fable 5 achieved 80.3% on SWE-Bench Pro (+11 points over Opus 4.8) and 29.3% on FrontierCode Diamond. It supports a 1M token context window and was priced at $10/M input and $50/M output tokens—half the price of Opus.
| Benchmark | Fable 5 | Opus 4.8 | GPT-5.5 |
|---|---|---|---|
| SWE-Bench Pro | 80.3% | 69.2% | 58.6% |
| FrontierCode Diamond | 29.3% | 13.4% | — |
| Spatial Reasoning | 38.6% | 14.5% | — |
Source: Vellum / TrueFoundry (June 2026)
Practical Impact on Agent Users and Recommended Alternatives
Developers using Fable 5 in Cursor or Claude Code for long-running tasks such as large-scale codebase migrations or complex workflow automation have suddenly lost access. Recommended alternatives include switching to Opus 4.8 or GPT-5.5, although these models score lower on benchmarks. Task decomposition and prompt optimization are necessary to mitigate the performance gap.
A real-world example previously cited was completing a 50M-line Ruby migration in a single day with Fable 5. With Opus 4.8, achieving comparable results may require multiple iterations and careful workflow design. API users should manually configure fallback to Opus 4.8.
Future Outlook and Anthropic’s Response
Anthropic is actively negotiating with the government and aims to restore limited access for trusted partners. The company plans to expand the Project Glasswing framework with enhanced safety evaluations before gradually reopening access. The Fable 5 API endpoint “claude-fable-5” will remain unavailable for the foreseeable future.
In the developer community, discussions are emerging around hybrid multi-model operations and migration to self-hosted agent frameworks to prepare for prolonged restrictions. Anthropic posted on X that it is “doing everything possible to minimize user impact.”
Frequently Asked Questions (FAQ)
Q1: What is the main difference between Fable 5 and Mythos 5?
Q2: Are existing Claude models affected by this regulation?
Q3: What alternatives do you recommend for agent tasks previously using Fable 5?
Q4: When is access expected to be restored?
Q5: What should API users do in response?
Related articles: Anthropic、Claude Fable 5 / Mythos 5 をリリース — 長時間・複雑タスク向け新世代モデル、Databricks、AI Agent向けメタハーネス「Omnigent」をオープンソース公開 — Claude Code / Codex横断でmulti-agent制御。
Author
krona23
Over 20 years in the IT industry, serving as Division Head and CTO at multiple companies running large-scale web services in Japan. Experienced across Windows, iOS, Android, and web development. Currently focused on AI-native transformation. At DevGENT, sharing practical guides on AI code editors, automation tools, and LLMs in three languages.
Leave a Reply