U.S. Clears Anthropic's Mythos 5 for Limited Use After Security Review

The U.S. Commerce Department has authorized Anthropic to restore access to its Mythos 5 AI model for select cybersecurity partners, resolving a two-week standoff that began when the government blocked all access over national security concerns.

Commerce Secretary Howard Lutnick informed Anthropic's chief compute officer that the model can now be released to "certain trusted partners," according to a letter dated Friday. The clearance follows intensive negotiations after the government abruptly barred foreign nationals from accessing both Mythos 5 and the related Fable 5 model on concerns that security guardrails could be circumvented.

What Changed

The government's original order on June 13 forced Anthropic to disable global access to both models. The restrictions represented the most significant federal intervention to date in an AI company's operations.

Mythos 5, Anthropic's most capable cybersecurity model, had been available to approximately 200 firms through Project Glasswing, an invite-only program. Participants included Apple, Google, Cisco, Nvidia, and Microsoft. The Commerce Department letter makes no mention of lifting restrictions on Fable 5, the public-facing version with additional safeguards.

Anthropic stated it "worked with the US government to address risks associated with the Covered Models," though the specific technical measures taken to satisfy security concerns were not disclosed. The company said conversations continue with the government to expand Mythos 5 access and restore availability of Fable 5.

The Jailbreak Problem

The government's action stemmed from concerns about "jailbreaking" — bypassing model guardrails. In response to the initial ban, Anthropic argued that "perfect jailbreak resistance is not currently possible for any model provider" and warned that applying such standards across the industry "would essentially halt all new model deployments for all frontier model providers."

Fable 5, released earlier this month as the first public version of the Mythos-class model, blocks certain cybersecurity and biology-related queries by routing them through Opus 4.8, a different model. Mythos 5 is a version of Fable 5 without some of these safeguards.

Industry Ripple Effects

The restrictions affected competitors beyond Anthropic. OpenAI limited the release of GPT-5.6 under pressure from the Trump administration, announcing Friday it would roll out a preview version only to government-approved partners before wider availability in coming weeks.

The standoff adds to ongoing tensions between Anthropic and the Trump administration. The Defense Department declared Anthropic a supply-chain risk in March after contract negotiations collapsed over the company's request for additional guardrails on military AI use.

What This Means

The partial clearance establishes a precedent for government oversight of advanced AI models with dual-use capabilities. By maintaining restrictions on Fable 5 while clearing Mythos 5 for limited deployment, the Commerce Department signals that even models with built-in safeguards may face federal scrutiny. The incident occurred weeks after Anthropic filed confidentially for an IPO at a valuation exceeding $900 billion, introducing regulatory uncertainty into the company's public market plans. The negotiations involved co-founder Tom Brown meeting directly with Commerce Secretary Lutnick, while CEO Dario Amodei took a hands-off approach, according to sources familiar with the talks.