Trump Administration Partially Lifts Ban on Anthropic’s Cybersecurity Models for Critical Infrastructure
Table of Contents
A Strategic Pivot in AI Security
Two weeks after a sweeping ban forced Anthropic to withdraw its most potent cybersecurity-oriented models from the market, the Trump administration has begun to walk back its restrictive stance. In a targeted shift, the government is now permitting the redeployment of Mythos 5 to a curated list of more than 100 U.S. government agencies and private sector companies.
The original restriction was designed to prevent foreign adversaries from leveraging high-end AI capabilities to develop sophisticated cyberattacks. However, the new directive acknowledges that the same tools are indispensable for those defending the nation’s digital perimeter. According to reports from Semafor and Reuters, the updated guidance allows these “trusted partners” to utilize the model, crucially including non-American employees within those organizations—a group previously blocked from access.
The Lutnick Directive and the ‘Trusted Partner’ Framework
The shift comes via a formal communication from Commerce Secretary Howard Lutnick to Anthropic’s chief compute officer, Tom Brown. In the missive, Lutnick stated he had determined that “appropriate safeguards are in place” to allow specific entities to access Mythos 5. This suggests a transition from a blanket ban based on nationality to a risk-based framework focused on the institutional legitimacy of the user.
Notably, the ban’s impact extended even to Anthropic’s own internal staff. Non-American employees at the AI lab were caught in the original dragnet, hindering the company’s ability to iterate on the model’s safety and performance. The latest move effectively restores the company’s internal operational capacity while expanding the model’s footprint across U.S. critical infrastructure.
The Fable 5 Limbo
While the restoration of Mythos 5 marks a significant victory for Anthropic, the situation remains complicated for Fable 5. Fable 5, a derivative of Mythos 5 released shortly before the ban, was marketed as having more robust safety guardrails. However, both models were pulled after independent security researchers demonstrated that these guardrails could be bypassed with relative ease.
The administration’s recent directive conspicuously omits any mention of Fable 5. While Mythos 5 is being returned to a controlled environment of vetted professionals, Fable 5 remains in a state of regulatory limbo. This distinction suggests the government views the high-risk, high-reward nature of Mythos 5 as a necessary tool for defense, while remaining skeptical of the “safe” version’s ability to resist adversarial attacks in a general-release setting.
Anthropic’s Response and the Path to General Access
Anthropic confirmed the development on Friday via a post on X, detailing their collaboration with federal authorities since June 12. The company characterized the move as a step toward restoring its “strongest cybersecurity model” to those responsible for operating and defending critical infrastructure.
For Anthropic, the priority now shifts to two fronts: the rapid restoration of access for the 100+ approved organizations and continued negotiations with the U.S. government to reopen Fable 5 for general public use. The incident highlights the precarious balance the administration is attempting to strike—fostering AI leadership and domestic defense while attempting to build an “AI firewall” against foreign intelligence services.
