# Mythos Partial Restore: From Hard Global Kill to Government-Licensed Islands of Access

Times of AI Desk · 2026-06-26 · Policy

[https://timesof.ai/2026/06/anthropic-mythos-5-partial-restore-trusted-partners](https://timesof.ai/2026/06/anthropic-mythos-5-partial-restore-trusted-partners)

> June 26 turns the Fable/Mythos crisis: Commerce approved Mythos 5 for ~100 trusted U.S. orgs while Fable stayed offline. Same day OpenAI gated GPT-5.6—frontier launches now pass through trusted-partner stages.

June 26 is the **turning point** of the Fable/Mythos crisis: from hard global kill to **government-licensed islands of access**. Combined with OpenAI’s same-day government-gated GPT‑5.6 preview, it crystallizes a new operating reality—frontier launches in the U.S. increasingly pass through **trusted-partner** stages rather than pure product calendars.

Two weeks after a Commerce export directive forced Anthropic to **globally disable** Claude **Fable 5** and **Mythos 5**, the U.S. government approved a **partial restore of Mythos 5** for a **set of trusted U.S. organizations**. Anthropic confirmed the June 26 approval in later primary posts; **Fable 5** stayed offline for the general public until the **June 30** full lift.

## What was restored on June 26

From Anthropic’s subsequent official timeline (redeploy posts referencing the June 26 approval), CNBC/Fortune/Politico reporting, and House-to-Commerce context:

- **Model**: **Claude Mythos 5** (cyber-focused / Glasswing-class access model).
- **Scope**: Restore for a **set of U.S. organizations**—described in contemporaneous reporting as on the order of **~100** vetted entities spanning critical-infrastructure defenders, federal agencies, and national labs (“trusted partners”).
- **Still restricted**: **Fable 5** public/commercial global access **not** restored on June 26; full Commerce lift came **June 30**, with Fable general availability **July 1**.
- **Process signal**: Commerce Secretary **Howard Lutnick** correspondence with Anthropic leadership (including COO **Tom Brown** in secondary accounts) framed conditional approval and reserved the right to **reevaluate** if commitments slipped.
- **Safeguards**: Anthropic messaging emphasized implemented fixes for the jailbreak concern that triggered the original order and ongoing work with peers on severity standards.

## Why partial, why Mythos first

Mythos was already a **controlled** defensive-cyber distribution path via **Project Glasswing**; restoring it to U.S. trusted partners was a lower-blast-radius step than flipping Fable back on for every Claude.ai user worldwide. The partial restore also addressed the **NSA capability gap** reported June 23 without immediately reopening foreign-national access fights.

## Limits

- ~100 trusted partners is contemporaneous press framing, not an Anthropic headcount filing.
- Fable remained offline for the public until June 30 / July 1—this is not a full restore day.
- Commerce reserved reevaluation rights if commitments slipped.

## Sources

- Anthropic: [“Redeploying Claude Fable 5”](https://anthropic.com/news/redeploying-fable-5) (anthropic.com/news/redeploying-fable-5) — primary post documenting June 26 Mythos restore approval and June 30 full lift.
- CNBC / Fortune / Politico reporting on June 26 Mythos clearance and trusted-partner framing.
- Anthropic X update referenced in company posts for the June 26 government approval.