Introduction

The U.S. government has partially reversed its earlier restrictions on Anthropic’s advanced AI models, allowing the company to release its powerful Mythos 5 system to over 100 “trusted” U.S. organizations. This decision, announced on Friday, marks a shift from a previous order to suspend access to the model due to national security concerns. The move comes amid heightened scrutiny of AI’s potential misuse by foreign adversaries and reflects the Trump administration’s ongoing efforts to balance innovation with oversight.

Anthropic, a leading AI developer, emphasized that the release of Mythos 5 is limited to entities critical to national security, such as those defending infrastructure and cybersecurity. The decision follows a broader regulatory framework aimed at curbing the risks of “frontier models” like Mythos and OpenAI’s GPT-5. However, the lack of transparency in selecting approved organizations has sparked criticism from experts and civil liberties advocates.

Background: The U.S. Government’s AI Oversight Framework

The U.S. government’s approach to regulating AI has intensified in recent years, driven by fears that advanced models could be weaponized by hostile nations or malicious actors. In June, the Trump administration issued an export control order halting access to Anthropic’s Mythos 5 and Fable 5 models, citing risks of misuse by foreign governments, particularly China and Russia. This followed a similar directive targeting OpenAI’s GPT-5, which the government urged to delay its public launch.

The regulatory framework is rooted in the National Defense Authorization Act and the Export Administration Regulations (EAR), which govern the export of dual-use technologies. The Trump administration’s recent executive order, signed in July, established a voluntary framework for AI developers to offer “covered frontier models” to the U.S. government for up to 30 days before releasing them to trusted partners. This initiative aims to ensure models are vetted for security risks before broader deployment.

Anthropic’s relationship with the U.S. government has been contentious. The company previously refused to allow the military to use its AI for domestic surveillance or autonomous weapons, leading to its inclusion on a national security blacklist. This history underscores the tension between innovation and control in the AI sector.

Main Developments: A Partial Reversal of Restrictions

The latest directive allows Anthropic to resume access to Mythos 5 for over 100 U.S. organizations, including many Fortune 500 companies. These entities are deemed critical to national security, such as those operating energy grids, financial systems, and defense infrastructure. The decision follows a six-week review of Anthropic’s collaboration with the government to address risks, including potential vulnerabilities in the model’s safeguards.

Key details include:
– Export License Waivers: Non-U.S. citizens working for approved companies or Anthropic employees will no longer need export licenses to access Mythos 5, though licensing restrictions remain for unapproved entities.
– Project Glasswing: Approved companies are part of Anthropic’s Project Glasswing, a coalition of 100+ tech firms and institutions. This initiative is seen as a way to align private-sector innovation with government priorities.
– Fable 5 Timeline: While Mythos 5 is being released, the government is also pushing for the eventual availability of Fable 5, a less restricted version of the model. However, no timeline has been set for its public release.

The move is part of a broader effort to ensure AI models are “trusted” before wider deployment. Anthropic stated that it is working with the government to expand access to Mythos 5 and eventually make Fable 5 available for general use.

Why This Matters: Balancing Innovation and Security

The decision to lift restrictions on Mythos 5 highlights the U.S. government’s dual priorities: fostering AI innovation while mitigating risks. However, the lack of transparency in selecting approved organizations has raised concerns about fairness and accountability.

National Security Concerns:
Mythos 5 is designed as a cybersecurity model, capable of identifying vulnerabilities in complex systems. If misused, it could accelerate sophisticated