The U.S. government has lifted a previously imposed export requirement that mandated Anthropic obtain a license before making its advanced AI models, Mythos and Fable, available internationally. This restriction had effectively restricted public access to these cutting-edge technologies, which are regarded as among the most sophisticated AI models released to date.
In a significant development for the AI community, Anthropic announced it would begin restoring access to these models starting Wednesday, July 1. The export restrictions were established on June 12 when the U.S. government categorized Mythos and Fable as export-restricted technologies. This classification meant that the models could not be accessed by foreign nationals without special permission, a logistical challenge that ultimately led Anthropic to cease public access entirely.
Following weeks of discussions, U.S. Secretary of Commerce Howard Lutnick highlighted that Anthropic had agreed to a series of commitments aimed at enhancing the security surrounding the models. These include proactively detecting and addressing security risks associated with Mythos and Fable, collaborating closely with the U.S. government to establish protocols and standards for future releases, and notifying the government of any malicious activities detected.
Prior to the introduction of the export rules, Anthropic had already committed to many of these security measures, raising skepticism among cybersecurity experts about the necessity and intent behind the restrictions. Critics suggested that these restrictions served more as a means of leveraging political power than providing any substantive security benefits, especially given the context of Anthropic leaders publicly voicing their concerns regarding the use of AI technology by the government.
Mythos was initially made available to a limited group of organizations starting in April, uniquely designed to address potential vulnerabilities within software systems. In June, Anthropic publicly launched a version of the model, Fable, which featured additional safety measures to guard against misuse.
With escalating competition from Asian AI firms launching models with capabilities comparable to Mythos—including Fugu and Tulonfeng—the U.S. government felt compelled to reconsider its stance to maintain competitiveness in the global AI landscape. Lutnick’s recent clearance enabled select customers, approved by the White House, to access Mythos. Additionally, OpenAI’s latest models have similarly been restricted to organizations chosen by the administration, rather than being publicly available.
The erratic nature of the Trump administration’s approach to AI policy has left many companies within the sector uncertain about future regulations governing AI model releases. An executive order released in June aimed at reviewing models before their distribution drew criticism from influential analysts, including Dean W. Ball, who has recently taken a policy role at OpenAI. This has created an atmosphere of ambiguity and concern regarding the future landscape of AI development and governance in the United States.
