US eases block on Anthropic’s Mythos 5, allowing limited release to key cyber defenders

In a significant policy shift, the US government has eased its stringent ban on Anthropic's advanced artificial intelligence model, Claude Mythos 5. This decision permits a limited number of American cybersecurity companies and essential infrastructure providers to utilize the model, marking a notable change just two weeks after the administration imposed strict export controls that halted access to Mythos 5 and another model, Fable 5, due to pressing national security concerns. Commerce Secretary Howard Lutnick communicated this development to Anthropic in a private letter, highlighting that thorough negotiations had paved the way for a regulated rollout of the technology. Lutnick assured that adequate safeguards are now in place for select trusted organizations to access the Claude Mythos 5 model. This initial phase will allow around 100 carefully vetted entities, primarily those involved in Anthropic’s defensive security initiative known as Project Glasswing, to immediately engage with the system, bypassing the usual export license requirements. The abrupt shutdown of access on June 12 sent shockwaves across the technology landscape, occurring mere hours after the public launch of the Mythos models. During preliminary testing, Mythos Preview showcased groundbreaking capabilities, including the ability to discover a 27-year-old vulnerability in OpenBSD’s TCP stack and to generate functional exploits for Mozilla’s Firefox browser at an unprecedented pace. The administration's intervention stemmed from concerns that adversaries such as China and Russia could exploit these advanced capabilities, posing threats to global banking and critical infrastructure systems. This recent reversal underscores the delicate balance the government is trying to maintain between ensuring national security and keeping American AI innovations competitive in a rapidly evolving global arena. This announcement coincided with OpenAI’s release of its own model, GPT-5.6, which is also undergoing a carefully controlled rollout. A spokesperson from the Commerce Department remarked on the diligence shown over the past two weeks to ensure that the US remains a leader in AI technology while also safeguarding its interests. Despite this progress for Anthropic, the broader landscape remains uncertain. The government’s announcement did not address Fable 5, which had briefly been available to the public before the restrictions were imposed. As of now, companies not on the approved list, as well as international partners and general consumers, remain in the dark about when they might gain access to this new technology. Anthropic has stated that it is collaborating with federal officials to establish long-term deployment standards and aims to restore access to Mythos 5, as well as make Fable 5 available to the general public once again.