Commerce Let Anthropic Restore Mythos to 100 Organizations

On June 26, Commerce Secretary Howard Lutnick told Anthropic it could turn Mythos 5 back on — but only for roughly 100 organizations on a federal allow list. Fable 5 stays offline for everyone else. Washington did not end the frontier-AI kill switch. It turned it into a guest list.

What's new: Lutnick's letter to Anthropic co-founder Tom Brown partially reversed the June 12 export-control order that forced Mythos 5 and Fable 5 offline worldwide — including for foreign nationals inside the United States.

• CNBC, which viewed the letter, quotes Lutnick: "I have determined that appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model."
• NBC News reports access returns to around 100 government agencies and private companies — many tied to Anthropic's Project Glasswing, including Cisco and JPMorgan Chase.
• The letter grants license-free access to entities in Annex A, their foreign-national employees, and Anthropic's own foreign-national staff — but says nothing about restoring Fable 5, the consumer-facing model.
• Semafor first reported the letter Friday night; Reuters and WIRED confirmed the partial reopening.

Why it matters: Frontier AI release is no longer a product decision. It is a federal clearance process with two tiers — approved critical infrastructure and everyone else.

What Anthropic said: On X Friday night, the company wrote: "Today, the government notified us that Mythos 5, our strongest cybersecurity model, can be redeployed to a set of US organizations that operate and defend critical infrastructure."

• A source close to Anthropic told NBC negotiations continue over the weekend to restore Fable 5.
• Co-founder Tom Brown — not CEO Dario Amodei — has reportedly led talks with the administration, per CNBC.
• Anthropic's supply-chain-risk designation from the Pentagon and its lawsuit to reverse the blacklisting remain active.

The same Friday, OpenAI complied: Hours before Lutnick's letter landed, OpenAI released GPT-5.6 Sol, Terra, and Luna to a "small group of trusted partners" at the government's request, per CNBC.

• OpenAI wrote: "We don't believe this kind of government access process should become the long-term default. It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them."
• CEO Sam Altman called the staggered debut "bad news" on X, per NBC — the company had planned a wider launch.
• An administration source told CNN the GPT-5.6 models are "on par" with Mythos — the same capability tier that triggered Anthropic's export ban.

How we got here: The June 12 shutdown followed a jailbreak report from an Amazon researcher, communicated through CEO Andy Jassy, that Fable 5's guardrails could be bypassed, per TechPolicy.Press.

• Because foreign nationals work at Anthropic and its partner firms, Lutnick applied export controls that forced a global shutdown — not just an overseas block.
• Anthropic noted the disclosed jailbreak affected models already on the market for months — not just Fable 5.
• Before the ban, Mythos had found thousands of cyber vulnerabilities; U.S. officials confirmed it identified flaws in classified systems within hours, per NBC and prior reporting.

The structural problem: TechPolicy.Press analyst Andrew Reddie writes that export law was built for goods crossing borders — not cloud APIs where the weights never leave U.S. servers but the capability does.

• Lutnick wrote that efforts since his June 12 letter had yielded "significant progress," per Reuters — and reserved authority to "reevaluate and adjust" license requirements and alter the Annex A list "at any time," per Semafor.
• Europe's tech chief Henna Virkkunen flew to Washington this week seeking Mythos access — Brussels signed Pax Silica the same day. The guest list does not include the EU.
• Legion, the ten-person legal-tech startup, is still suing Commerce over the original shutdown — proof that contracts mean nothing when export law treats a hosted model as a controlled export.

What to watch:

• Whether Fable 5 returns this weekend or becomes a permanent second-class tier behind Mythos.
• Who is on Annex A — and whether the list leaks or becomes the de facto frontier-AI procurement standard.
• Whether OpenAI's "couple of weeks" public rollout holds — or the trusted-partner preview becomes the real release schedule.

Convina's view: Lutnick did not solve the Mythos crisis. He managed it. One hundred organizations get the cyber model Washington spent two weeks fighting over; every other buyer — including paying enterprise customers and the general public — waits outside Annex A. That is not safety governance. It is rationing by political clearance. The labs racing toward IPO cannot pitch infinite scale while Commerce holds the customer list and the kill switch in the same drawer.