Fourteen days after the US government shut down both models, Commerce Secretary Lutnick sent a letter on 26 June restoring Mythos 5 for 100+ critical infrastructure organisations. Fable 5 is offline with no restoration date. The same day, OpenAI launched GPT-5.6 to only 20 government-approved partners. A new era of frontier AI governance just arrived.
The Mythos 5 is partially restored — fourteen days after the US government suspended both Claude Fable 5 and Claude Mythos 5 under a national security export control directive.
Anthropic said that the US government has allowed it to release its powerful Claude Mythos 5 model to some trusted US organisations, partially reversing an order two weeks ago to suspend access over national security risks. More than 100 companies and institutions will have access to Mythos 5, including many Fortune 500 companies. “Today, the government notified us that Mythos 5, our strongest cybersecurity model, can be redeployed to a set of US organisations that operate and defend critical infrastructure,” Anthropic said in a statement on Friday.
By contrast, Claude Fable 5 — the general-release version of the same underlying model — is offline for all users globally. Anthropic confirmed it serves zero Fable 5 traffic. No restoration date has been announced.
On the same day, OpenAI launched GPT-5.6 Sol to just 20 government-approved partners — the first ever US government-gated AI model release. The frontier AI governance landscape changed fundamentally in a single week.
What’s Happening & Why It Matters
The Lutnick Letter — Authorisations and Restrictions
The Mythos 5 partial restoration rests on a specific legal instrument. Commerce Secretary Howard Lutnick wrote a letter to Anthropic on 26 June granting partial restoration authority. Lutnick said in the letter that an export licence will no longer be needed for Mythos 5 to trusted companies and their employees who are not US citizens, or to Anthropic’s employees who are not US citizens, but licensing restrictions are in place for companies that are not on the approved list.
That language is precise and deliberate. The Lutnick letter does not lift the export control directive. It creates an approved list of organisations that are outside it. Companies on the list — most of them Project Glasswing partners — can access Mythos 5 for their non-US citizen employees without an individual export licence. Companies not on the list are subject to the original June 12 directive. The list is not public. “No one knows how companies are picked and why everyone else is excluded,” said John Coleman, legislative counsel at the Foundation for Individual Rights and Expression.
Fable 5: Still Offline — and the ID Verification Path Being Built
The Mythos 5 partial restoration resolves the critical infrastructure track. It leaves the general public track entirely unresolved. Anthropic confirmed on 25 June that Anthropic staff member Sam McAllister stated the company was “currently serving exactly 0 traffic to Fable 5.” Viral reports of a staged Fable 5 return via Claude Code were confirmed as incorrect — a front-end UI bug showing Fable 5 in the model picker from historical context, not actual access.
By contrast, a restoration mechanism is emerging. Anthropic updated its privacy policy effective 8 July 2026 to require government-issued ID and biometric verification through identity provider Persona. That verification infrastructure appears to be the mechanism that will allow Fable 5 to return for verified US citizens while maintaining the export control restriction on non-US nationals. The framework is not yet operational. Additionally, the government is moving towards allowing Fable 5’s release, though a timeline is unclear.
The Korean Telecom Trigger — and 28.8 Million Fake Accounts
The Mythos 5 partial restoration follows the disclosure of what actually triggered the original June 12 shutdown. Anthropic Managing Director of International Chris Ciauri confirmed at a Seoul press conference that a Korean telecommunications company with access to Claude Mythos triggered the US government’s decision. That company apparently used Mythos in a way that prompted the national security concern — though Ciauri declined to provide details.
A second irony emerged in Washington coverage. Anthropic‘s own 10 June Senate letter — the Alibaba distillation accusation TF covered in its Anthropic Alibaba article — documented 25,000 fraudulent accounts running 28.8 million Fable 5 exchanges undetected between April and June 5. Critics ask how a model seen as uniquely dangerous for cybersecurity could simultaneously miss industrial-scale account farms using dynamic IPs and reseller infrastructure. Anthropic‘s answer distinguishes between export controls targeting foreign-national API access and bot detection targeting fake domestic accounts — two different enforcement mechanisms addressing two different threat categories.
The House Homeland Security Demo — Garbarino’s Alarm
The Mythos 5 partial restoration coincides with significant Congressional attention. House Homeland Security Committee chair Representative Andrew Garbarino hosted an Anthropic Fly Out Day demo at which Mythos was instructed to find a vulnerability in a bank system — and demonstrated a pathway to both exploit and patch it. Punchbowl News reported Garbarino called the capability “alarming” and said roughly 95% of Congressional colleagues do not grasp the implications. Additionally, Garbarino aligned himself with the pre-release government access framework established in the 2 June executive order — arguing the government needs early access before any public release of Mythos-class models.
That Congressional view shapes the Fable 5 restoration negotiation. The more legislators who see Mythos capability demonstrations, the harder it is politically to argue that general public access should be unrestricted.
GPT-5.6 Sol: The First Government-Gated AI Model Release
The second story from 26 June is arguably more consequential in the long run. OpenAI unveiled GPT-5.6, a three-model suite led by its flagship Sol, but in an unprecedented move the company restricted initial access to roughly 20 organisations individually approved by the US government. The launch is the first time an American AI company has released a frontier model under a government-managed access list.
The three-model structure is specific. Sol is the flagship at $5 per million input tokens. Terra is at $2.50 per million. Luna is at $1 per million. Sol scored 96.7% on OpenAI’s internal Capture-The-Flag cybersecurity evaluations, crossing what the company classifies as a ‘high’ cyber risk threshold. By contrast, OpenAI was unambiguous that the arrangement is not sustainable. “We’ve made clear to the US government that this is not our preferred long-term model, and will work with them and others in industry to achieve a more sustainable approach for future releases,” Altman wrote in an internal memo.
The New Template — What Both Releases Establish
The Mythos 5 partial restoration and the GPT-5.6 Sol government-gated release together establish something that did not exist a month ago: a de facto pre-release government review process for frontier AI with advanced cybersecurity capabilities. The White House’s forced shutdown of Anthropic’s Mythos model earlier in June appears to have set the operational template. Government review precedes availability, and Washington decides who gets in first.
The Trump administration’s June 2 executive order called on AI companies to voluntarily share frontier models with the government up to 30 days before public release. In practice, neither the Anthropic nor the OpenAI situation was voluntary — both companies faced commercial consequences for non-compliance that made the distinction between voluntary and mandatory largely theoretical. As TF covered in its Trump AI security order article, the word “voluntary” in the June 2 executive order carried the ambiguity from the moment it was signed.
TF Summary: What’s Next
Mythos 5 access is being provisioned to the 100+ approved organisations immediately. Fable 5 restoration is pending — the government is moving towards approval but no date is confirmed. Anthropic‘s ID verification requirement via Persona takes effect 8 July 2026. OpenAI‘s GPT-5.6 expands to more approved partners next week, with a general release targeted “a couple of weeks” after the limited preview period. The federal benchmarking and assessment process for frontier AI models targets an August 2026 deadline.
MY FORECAST: The Mythos 5 partial restoration is the clearest signal yet that the US government’s frontier AI governance framework will become permanent — not temporary. The Lutnick letter’s approved-list mechanism, Anthropic‘s ID verification rollout, and OpenAI‘s government-gated GPT-5.6 launch together constitute the first operational version of the pre-release review system that the June 2 executive order described in skeleton form. By contrast, Fable 5‘s general release will return — probably within two to three weeks, once the ID verification infrastructure is operational and the government confirms the jailbreak concern is resolved. The more durable consequence is structural. Every frontier AI lab developing models with advanced cybersecurity capabilities understands that commercial launch requires government coordination. Google DeepMind, Meta AI, xAI, and Mistral are all building in the environment. The governance template Anthropic and OpenAI experienced in June 2026 is the template the next generation of frontier labs inherits.
Related Stories
- Anthropic Pulls Claude Fable 5 — Four Days After Launch, on Government Order
- Anthropic Accuses Alibaba of 28.8 Million Fake Queries to Clone Claude
- Trump Signs AI Security Order — Scaled Back, Voluntary, 12 Days Late