NSA Using Anthropic's Mythos Despite Defense Department Blacklist

The National Security Agency (NSA) has been using Anthropic's AI model Mythos for intelligence analysis despite the Defense Department placing it on a procurement blacklist, according to sources familiar with the matter.

This apparent contradiction in federal AI procurement practices highlights the complex and sometimes conflicting approaches different government agencies take when adopting artificial intelligence tools for national security purposes.

The Blacklist Context

The Defense Department's blacklist of certain AI models, including Mythos, was implemented last year amid growing concerns about data security, foreign influence, and the reliability of commercial AI systems for sensitive military applications. The blacklist specifically targeted models from companies that hadn't undergone rigorous security vetting or that posed potential risks to classified information.

Anthropic, founded by former OpenAI researchers and backed by major investors including Google and Amazon, has positioned itself as a safety-focused AI company. However, the Defense Department's concerns appear to center on the model's training data origins and potential vulnerabilities rather than the company's stated safety commitments.

NSA's Position

NSA officials have defended their use of Mythos, arguing that the agency's specific use cases for intelligence analysis don't trigger the same security concerns that led to the Defense Department's restrictions. The NSA claims its implementation includes additional security layers and operational controls that mitigate the risks identified by the Pentagon.

"The NSA's mission requirements are distinct from those of the broader Defense Department," said a senior intelligence official who spoke on condition of anonymity. "Our use of Mythos is carefully controlled and serves specific analytical functions that we've determined to be low-risk."

What This Means

This situation reveals the fragmented nature of AI governance within the U.S. government. While the Defense Department has taken a cautious, centralized approach to AI procurement, other agencies like the NSA appear to be pursuing more flexible strategies that prioritize operational effectiveness over blanket restrictions.

Industry analysts suggest this could set a precedent for how other government agencies approach AI adoption, potentially creating a patchwork of different standards and security protocols across federal agencies.

"What we're seeing is the tension between security caution and operational necessity," said Dr. Sarah Chen, an AI policy researcher at Georgetown University. "The NSA clearly believes the intelligence value of Mythos outweighs the theoretical risks, while the Defense Department has taken a more conservative stance."

Broader Implications

The NSA's decision to use Mythos despite the blacklist raises questions about the effectiveness of government-wide AI procurement policies. If agencies can simply bypass restrictions by claiming unique mission requirements, it could undermine efforts to establish consistent security standards for AI systems handling sensitive government data.

This case also highlights the growing influence of commercial AI companies in government operations, even when their products haven't received full security clearance. As agencies compete for AI capabilities to maintain technological advantages, the pressure to adopt cutting-edge tools may increasingly conflict with security protocols.

Anthropic has not publicly commented on the NSA's use of Mythos or the Defense Department's blacklist. The company maintains that its models undergo rigorous safety testing and that it works closely with government partners to ensure appropriate security measures are in place.

The situation underscores the ongoing challenge of balancing innovation and security in government AI adoption, a debate that's likely to intensify as artificial intelligence becomes increasingly central to national security operations.

Photo: Samyukta Lakshmi/Bloomberg via Getty Images