NSA Using Anthropic's AI Despite Supply Chain Risk Designation

The National Security Agency and Department of Defense are using Anthropic's Mythos Preview AI system despite the government's designation of Anthropic as a supply chain risk, according to multiple sources familiar with the matter.

In February, the Department of Defense moved to cut off Anthropic and force its vendors to follow suit, citing security concerns about the AI company's operations. However, sources tell Axios that both the NSA and DoD have continued using Mythos Preview internally.

This creates an unusual situation where government agencies are simultaneously restricting Anthropic while relying on its technology for sensitive operations. The contradiction highlights the challenges agencies face in balancing security protocols with the practical needs of advanced AI capabilities.

The use of Anthropic's technology within intelligence and defense circles comes as the company faces scrutiny over its supply chain practices. The designation as a supply chain risk typically triggers restrictions on government use and procurement, yet operational needs appear to be overriding these concerns.

This isn't the first time government agencies have found themselves in a bind over AI security designations. Similar situations have occurred with other tech companies where the strategic value of their technology conflicted with security assessments.

The revelation raises questions about the effectiveness of supply chain risk management in the AI era, where the line between commercial and national security applications continues to blur. As agencies grapple with these competing priorities, the enforcement of security designations may become increasingly complicated.

For Anthropic, the continued use by government agencies despite the official ban suggests their technology remains valuable enough to justify the security trade-offs. The company has not publicly commented on the apparent contradiction between the government's official stance and its actual usage patterns.

This situation underscores the broader tension in government AI adoption: agencies need cutting-edge capabilities to maintain technological superiority, but those same capabilities often come with security risks that are difficult to fully mitigate.