Anthropic Cleared for Limited Release of Mythos AI

In a significant development for the burgeoning field of artificial intelligence, the U.S. government has granted Anthropic, a leading AI safety and research company, provisional approval to release its advanced Mythos 5 model. The authorization permits access for a select group of approximately 100 companies and federal agencies, a move confirmed by CNBC. This decision represents a crucial advancement in the ongoing dialogue between the current administration and Anthropic, which had been navigating a two-week impasse over its latest generative AI offerings, Fable 5 and Mythos 5.

In a formal letter to Anthropic, the Department of Commerce outlined the conditions for this phased rollout. U.S. Commerce Secretary Howard Lutnick stated, “I have determined that appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model.” This directive, while specific to Mythos 5, did not extend to the Fable 5 model, indicating a continued point of contention or a need for further evaluation of that particular offering.

The situation arose after Anthropic proactively suspended access to both Fable 5 and Mythos 5 earlier this month. This action was in response to a government directive concerning export controls, citing “national security authorities.” The company had indicated that it was instructed to cease all access for foreign nationals, both within and outside the United States, including its own international employees. This decision underscored the delicate balance between fostering AI innovation and mitigating potential national security risks.

The announcement of Mythos 5 and Fable 5 just days prior to this suspension had positioned them as state-of-the-art models, excelling across various industry benchmarks. Fable 5, in particular, was hailed as Anthropic’s most sophisticated public release to date, incorporating novel safety features designed to restrict responses in sensitive or high-risk domains. The inclusion of these safeguards was a direct response to the evolving regulatory landscape and the increasing emphasis on responsible AI deployment.

This development echoes a similar cautionary approach taken by a key competitor. Earlier on Friday, OpenAI, another prominent AI developer, announced the release of three new AI models – GPT-5.6 Sol, Terra, and Luna. In line with governmental recommendations, OpenAI is initially limiting the availability of these new models to a “small group of trusted partners.” The company expressed its commitment to broad accessibility, with plans to make these models generally available in the coming weeks, emphasizing that they had proactively shared their capabilities and rollout strategies with the government.

Anthropic’s engagement with the current administration has, however, been more complex. Earlier this year, the Department of Defense had engaged in a significant dispute with Anthropic regarding the use of its AI models. Following unsuccessful negotiations, the DoD designated Anthropic as a supply chain risk, a classification typically reserved for entities posing a threat to U.S. national security. This designation necessitates defense contractors to certify that Anthropic’s Claude models are not incorporated into their military projects.

In response to this designation, Anthropic initiated legal action against the administration, seeking to overturn the blacklisting. This litigation remains ongoing, highlighting the intricate legal and strategic challenges faced by AI companies operating at the cutting edge of technology and national policy.

The approval for Mythos 5 represents a potential turning point, signaling a path forward for Anthropic to engage with critical government entities and trusted industry partners. The careful, phased rollout, coupled with ongoing dialogue, suggests a growing understanding of the need for robust AI governance that balances innovation with security. As the AI landscape continues its rapid evolution, such collaborative approaches will be crucial in shaping the future of this transformative technology.