Policy

NSA Using Anthropic's Mythos Preview Despite Pentagon Supply Chain Risk Blacklist

On April 19, 2026, reporting revealed that the National Security Agency is actively using Anthropic's most powerful AI model, Mythos Preview, for cybersecurity tasks such as vulnerability scanning, even as the Department of Defense — which oversees the NSA — has designated Anthropic a 'supply chain risk' and is seeking to block its use across military contractors.

Tech Insights Reporter Apr 19, 2026 7 min read Washington, D.C.

title: "NSA Using Anthropic's Mythos Preview Despite Pentagon Supply Chain Risk Blacklist" summary: "On April 19, 2026, reporting revealed that the National Security Agency is actively using Anthropic's most powerful AI model, Mythos Preview, for cybersecurity tasks such as vulnerability scanning, even as the Department of Defense — which oversees the NSA — has designated Anthropic a 'supply chain risk' and is seeking to block its use across military contractors." category: "Policy" author: "Tech Insights Reporter" date: "2026-04-19" readTime: "7 min read" location: "Washington, D.C." hue: 260 regions: - "us" tag: "AI Policy" featured: false

TLDR

Axios reported on April 19, 2026, that the NSA is using Anthropic's Mythos Preview — a frontier model with advanced cyber capabilities — despite the Pentagon's formal designation of Anthropic as an "unacceptable supply chain risk." The model, restricted by Anthropic to roughly 40 organizations due to its offensive potential, is being deployed for defensive purposes like scanning networks for vulnerabilities. This comes amid an ongoing legal battle where the DoD is enforcing the blacklist on vendors and contractors, while intelligence community needs appear to be driving continued access. The revelation highlights tensions between U.S. government branches on AI adoption, safety, and national security priorities.

The Blacklist and Operational Use

In February 2026 (with public details emerging around March), the Department of Defense designated Anthropic a supply chain risk. This followed contract disputes where the Pentagon sought broader access to models for "all lawful purposes," including potentially mass surveillance and autonomous weapons development. Anthropic resisted, citing safety and ethical boundaries.

Key details:

The designation aims to restrict Anthropic tools from DoD contractors and vendors.
A court temporarily blocked aspects of the "Orwellian" measure.
Despite this, the NSA — part of the DoD — is using Mythos Preview.
Sources indicate the model is primarily for finding and exploiting vulnerabilities in the agency's own systems.
Broader DoD use was also reported by some sources.

Anthropic has limited Mythos Preview access to about 40 organizations (only 12 publicly announced under Project Glasswing for defensive cyber collaboration with firms like Microsoft, Google, and others). The model has demonstrated capabilities to identify thousands of vulnerabilities across major OSes and browsers, including long-unpatched ones, and chain exploits effectively.

Context of the Feud and Government Dynamics

The conflict stems from earlier 2026 contract renegotiations. DoD officials argue Anthropic's restrictions prove it cannot be fully trusted in wartime scenarios. Anthropic maintains its stance protects against misuse while advancing defensive AI.

Anthropic CEO Dario Amodei met with White House officials (including chief of staff Susie Wiles and Treasury Secretary Scott Bessent) around April 17 to discuss government use of Mythos and security practices.
The UK’s AI Security Institute has also received access.
The disclosure underscores fragmentation: intelligence agencies prioritize operational tools, while defense procurement enforces strict risk controls.
Contractors must certify non-use, creating compliance challenges.

This is not isolated; similar tensions exist with other frontier labs as models gain cyber potency.

Why this story matters

The April 19, 2026 reporting exposes a core inconsistency in U.S. AI governance: the same department (DoD) that blacklists a company for supply chain risks oversees an agency (NSA) actively deploying its most powerful model for critical cyber defense. With Mythos limited to a small trusted group due to its dual-use potential, the story illustrates how national security imperatives are accelerating adoption of frontier AI even amid legal and policy battles. It raises questions about coherence in export controls, contractor compliance, intelligence-defense coordination, and the challenges of governing tools that can both secure and threaten critical infrastructure. As frontier models' capabilities outpace policy frameworks, such internal contradictions may define the next phase of responsible AI deployment in government.

Sources

Axios: “Scoop: NSA using Anthropic's Mythos despite Defense Department blacklist” (April 19, 2026). https://www.axios.com/2026/04/19/nsa-anthropic-mythos-pentagon
Reuters: “US security agency is using Anthropic's Mythos despite blacklist, Axios reports” (April 19, 2026).
Cross-referenced coverage from TechCrunch, Engadget, and prior reporting on the DoD designation, Project Glasswing, and contract disputes (February–April 2026).

Featured Image Alt Text

Illustration of U.S. government agencies with conflicting policies on Anthropic's Mythos AI model: NSA using advanced cyber tools while Pentagon enforces blacklist, with digital shields, vulnerability scans, and policy documents dated April 19 2026