Mythos 5 Is Back — for 100 US Companies. Fable 5 Still Dark.

US Commerce Dept authorized Mythos 5 re-release to ~100 vetted US orgs on June 26, reversing the June 12 export-control suspension. Project Glasswing companies get access. Fable 5 remains blocked. Government now effectively picks who gets frontier AI.

Key facts: On June 26, 2026, US Commerce Secretary Howard Lutnick authorized Anthropic to re-release Claude Mythos 5 to approximately 100 pre-vetted US organizations — partially reversing the June 12 export-control suspension that killed access to both Mythos 5 and Fable 5. Approved companies include Fortune 500 firms and critical infrastructure providers, many from Project Glasswing. Fable 5 remains fully blocked with no timeline. Foreign nationals working for approved companies no longer need individual export licenses. The government is now effectively the gatekeeper for frontier AI model access in the US.

What happened

On June 12, Commerce Secretary Lutnick invoked export-control authority to force Anthropic to disable Mythos 5 and Fable 5 entirely — even for US customers — because foreign-national employees at partner companies could theoretically access the models. (Source: Anthropic official statement, June 12, 2026.)

Two weeks later, on June 26, Lutnick sent a letter stating Anthropic had made “significant progress” addressing the government’s concerns and authorized a limited re-release of Mythos 5.

What’s back: Mythos 5 for ~100 approved US organizations. No individual export licenses needed for their foreign-national employees.

What’s still dark: Fable 5 — the consumer/prosumer-facing model — remains fully suspended. No timeline announced.

What’s unclear: Whether non-US partners (Samsung, SK Hynix, SK Telecom, KISA — all previously in Project Glasswing) regain access. Early reports suggest Korean access “remains unclear.” (Source: Aju Press, June 27, 2026.)

Project Glasswing: the approved list

Many of the 100 approved organizations are part of Project Glasswing, Anthropic’s invite-only cybersecurity initiative. Before the shutdown, about 200 firms had access, including Apple, Google, Cisco, Nvidia, Microsoft, and JPMorgan Chase.

Project Glasswing uses frontier AI to identify and patch software vulnerabilities in critical infrastructure. The program’s defensive cybersecurity focus is likely why Mythos 5 — described as Anthropic’s “strongest cybersecurity model” — was prioritized for re-release over Fable 5.

The government as gatekeeper

This is the most significant government intervention into AI model access to date. The Commerce Department now effectively decides which companies can use the most powerful AI models — without a transparent application process.

OpenAI CEO Sam Altman criticized the approach: “I just don’t like the idea of the government picking the customers.” (Source: Reuters, June 26, 2026.)

John Coleman of FIRE called it “putting too much power in the hands of the government” with no transparency.

The broader framework: Trump’s executive order establishes a voluntary process where AI developers offer “covered frontier models” to the government for up to 30 days of review before releasing them to trusted partners. Both Anthropic (Mythos 5/Fable 5) and OpenAI (GPT-5.6) are now operating under this de facto regime.

Anthropic continues negotiating to expand Mythos 5 access and restore Fable 5. The company was pursuing an IPO with a valuation topping $900 billion at the time of the suspension. (Source: Bloomberg, June 2026.)

What this means if you’re building with Claude models

If your company is on the list: You now have API access to Mythos 5 again. Verify with your Anthropic account rep and re-enable any pipelines that were pinned to claude-mythos-5.

If not: You’re still blocked from Mythos 5. Use claude-opus-4-8 — Anthropic designed it as the explicit fallback for flagged requests. Opus 4.8 is production-stable, widely available, and $5/$25 per 1M tokens (half of Mythos 5 pricing).

Model-swappable architecture is now essential. The events of June 12–26 prove that frontier model access can be revoked overnight by government action — and restored selectively. Your agent loops and coding pipelines should reference models by config flag, not hardcoded strings. A model that’s available today may be dark tomorrow; a model that’s dark today may be back next week for “approved” customers only.

The tiered-access regime is here. We now have three tiers: government-approved enterprise (Mythos 5), public API (Opus 4.8), and fully blocked (Fable 5). If you’re building a product that relies on frontier model capabilities, understand which tier your customers fall into — and build fallback chains accordingly.

Monitor the Fable 5 timeline. When Fable 5 returns (and it will — the Commerce letter explicitly mentions ongoing negotiations), it will likely follow the same “trusted partners first” pattern as Mythos 5. Have your model config ready to swap.