The US government has eased restrictions on Anthropic’s advanced Mythos AI model, allowing the company to restore access for a limited group of trusted organizations after addressing cybersecurity concerns that prompted an export suspension earlier this month. The move reflects a broader effort to balance national security with continued progress in artificial intelligence as policymakers grapple with increasingly capable AI systems.
Government Revises Export Restrictions
According to a letter from Commerce Secretary Howard Lutnick, Anthropic worked closely with federal officials to reduce risks associated with its most advanced AI models. After reviewing the company’s proposed safeguards, the Commerce Department concluded that selected cybersecurity firms and critical infrastructure providers could regain access to the Claude Mythos 5 model.
The approval remains narrowly targeted. The government has not yet authorized the broader release of Fable 5, a less powerful version of the Mythos family, and discussions over expanding access are still underway.
Anthropic Restores Access for Trusted Partners
Anthropic confirmed that it has received approval to redeploy Mythos 5 to a small group of vetted cyber defenders and infrastructure organizations. The company said it is moving quickly to restore service while continuing discussions with federal officials on expanding access and eventually making Fable 5 available again.
Earlier this month, Anthropic suspended customer access to both Mythos and Fable after complying with government instructions to halt use by foreign nationals. The restrictions even affected some of the company’s own employees, highlighting the unusually broad scope of the export controls.
AI Security Debate Intensifies
The temporary ban has drawn attention to the lack of a consistent regulatory framework for frontier AI models. Although the US has generally favored limited AI regulation, recent actions suggest officials are taking a more cautious approach to systems capable of accelerating cybersecurity research and vulnerability discovery.
The concerns are not without economic consequences. IBM’s Cost of a Data Breach Report 2024 found that the global average cost of a data breach reached a record $4.88 million, underscoring why governments are paying closer attention to technologies that could strengthen either cyber defense or cyber offense.
Cybersecurity expert Bruce Schneier has long argued that powerful technologies are rarely inherently good or bad. Their real-world impact depends on the safeguards, governance, and incentives surrounding their deployment, a perspective that aligns with the government’s measured approach to advanced AI models.
Anthropic’s relationship with Washington has also faced previous challenges. Earlier this year, the company was designated a supply chain risk following disagreements over military applications of its technology. Anthropic challenged that decision in court and has already secured an early legal victory during the ongoing proceedings.
GrowBusinessMag notes that the latest decision illustrates how AI oversight is shifting from broad restrictions toward controlled deployment with defined safeguards. As talks continue between Anthropic and US officials, the next milestone will be determining whether Fable 5 can return to wider availability under a more predictable regulatory framework.
