US Government Lifts Block on Anthropic's Claude Mythos 5 AI Model, Allows Limited Release to Select American Organizations

Jun 26, 2026

The US government has partially lifted its export ban on Anthropic's Claude Mythos 5 AI model, allowing a limited release to over 100 approved American organizations and government agencies. Commerce Secretary Howard Lutnick confirmed that appropriate safeguards are now in place for defensive cyber use. The decision marks a de-escalation after a June 12, 2026, block sparked by national security fears. Meanwhile, rival OpenAI has also agreed to a government-requested staggered release of its new GPT-5.6 models, highlighting a broader shift toward active federal vetting of frontier AI.

Mythos 5 limited release authorization

▪The United States government authorized Anthropic on June 26, 2026, to redeploy its Claude Mythos 5 AI model to a select group of trusted American organizations.
▪The limited release of Claude Mythos 5 will grant access to more than 100 US organizations, including private companies, Fortune 500 corporations, and government agencies.
▪US Commerce Secretary Howard Lutnick sent a letter to Anthropic Chief Compute Officer Tom Brown stating that appropriate safeguards are in place to permit trusted partners to access Claude Mythos 5.
▪The new directive allows approved organizations and Anthropic to let their foreign national employees access the Claude Mythos 5 model without requiring an export license.

Export control national security invocation

▪The Trump administration cited concerns that powerful AI systems could be misused by military intelligence users in China, Russia, or other foreign countries.
▪The Trump administration grew concerned about Anthropic's rollout of Mythos after learning the company granted access to a South Korean telecommunications firm believed to have ties to China.
▪US Commerce Secretary Howard Lutnick invoked export control authorities on June 12, 2026, to block access to Anthropic's Claude Mythos 5 and Fable 5 models over national security concerns.

Project Glasswing cybersecurity applications

▪An early version of Claude Mythos 5 demonstrated the capability to find thousands of new cyber vulnerabilities and bugs.
▪Prior to being pulled, Claude Mythos 5 was available to a subset of trusted organizations in Anthropic's Project Glasswing, including Cisco and JPMorgan Chase.

OpenAI GPT-5.6 phased release

▪OpenAI CEO Sam Altman expressed dissatisfaction with the staggered release of GPT-5.6, calling the government-mandated customer-by-customer approval process bad news.
▪OpenAI announced on June 26, 2026, that it is staggering the release of its latest GPT-5.6 series of models in phases at the request of the US government.

Trump AI safety executive order

▪The Trump administration's recent interventions represent a shift from its previous deregulatory stance on artificial intelligence.
▪President Donald Trump signed an executive order in June 2026 directing the federal government to establish a safety testing and vetting mechanism for advanced AI models.

Fable 5 guardrail circumvention concerns

▪The US government's latest directive does not authorize the release of Fable 5, though Anthropic is continuing talks with the government to restore its access.
▪Amazon and the National Security Agency raised concerns to the White House that Anthropic's Fable 5 model could be easily jailbroken by malicious actors.