Pentagon Uses Anthropic Mythos AI for Cybersecurity

Pentagon Deploys Anthropic’s Mythos AI to Defend US Government Networks

The United States Department of Defense has deployed Anthropic‘s Mythos cybersecurity model across government networks. The model targets dangerous software vulnerabilities that have existed for decades. Officials confirmed the deployment is currently underway. The move marks a significant moment in the intersection of artificial intelligence and national defense.

The Defense Department‘s top technology official confirmed the news on Tuesday. Emil Michael, the Pentagon’s chief technology officer, described the situation as a national security moment. He said the government has hardened its networks using Mythos. The model carries unique capabilities specifically suited to finding and patching cyber vulnerabilities.

Mythos was formally announced on April 7. Anthropic deployed it under a controlled initiative called “Project Glasswing.” Under this program, select organizations gain access to the unreleased Claude Mythos Preview model. Access is strictly limited to defensive cybersecurity purposes.

Anthropic states that Mythos can identify long-standing flaws in web browsers, critical infrastructure, and widely used software. The model scans systems for weaknesses that have gone undetected for years. Its detection capabilities extend across multiple layers of digital infrastructure. This makes it a powerful tool for government-level defense operations.

A National Security Moment Driving Urgent Action

Michael addressed attendees at a conference in Washington, DC. He explained that these vulnerabilities have always existed within government systems. What has changed is the speed at which AI tools can now detect them. Those same tools, however, also allow attackers to exploit those weaknesses faster.

This creates a sharp tension at the heart of AI-driven cybersecurity. The capabilities that protect networks can also become weapons against them. Defenders and attackers now operate in the same accelerated environment. Michael made clear that the government views speed as the deciding factor here.

The Pentagon’s use of Mythos reflects a calculated strategic decision. Officials chose to harness the model’s defensive power despite significant political complications. The deployment continues even as the Defense Department works to remove Anthropic‘s products from its broader operations. Michael confirmed the Pentagon plans to complete that transition in the coming months.

The coexistence of these two positions is striking. The government simultaneously deploys Anthropic technology and moves to eliminate it. Officials frame this as a pragmatic response to an urgent threat. The defensive value of Mythos, they argue, outweighs the complications of its use.

Supply-Chain Risk Label Shadows the Deployment

The Defense Department previously declared Anthropic a supply-chain risk. The two sides failed to reach an agreement on how Anthropic’s AI models could be used by the agency. That disagreement led directly to the Pentagon’s formal blacklisting of the company. The relationship between the two organizations remains deeply strained.

Anthropic responded to the Pentagon’s blacklisting decision by taking legal action against the Trump administration in March. The company sued in an attempt to reverse the Pentagon’s blacklisting. That legal battle continues while Mythos remains in active government use. The situation underscores the unprecedented complexity of AI procurement in the national security space.

Despite the blacklisting, officials chose to proceed with Mythos under Project Glasswing. The controlled nature of the program provides a structured framework for use. Only select organizations receive access, and usage remains confined to defense. This framework appears to have given officials the confidence to move forward.

Anthropic’s Window Is Temporary, Rivals Are Closing In

Emil Michael was candid about Anthropic‘s current competitive advantage. He stated plainly that the advantage is temporary. Rival models from OpenAI, xAI, and Google will arrive soon. When they do, the cybersecurity landscape will shift again.

This acknowledgment carries significant implications for defense strategy. The government cannot afford to become dependent on a single vendor’s capabilities. Michael‘s statement signals awareness of that risk. The Pentagon appears to be planning for a future with multiple competitive AI options in cybersecurity.

OpenAI, xAI, and Google each bring formidable resources to the AI security space. Their entry into this specific capability domain could rapidly change the calculus. For now, Anthropic holds a distinct edge with Mythos. But officials are already preparing for that edge to erode.

Broader Stakes for AI in National Defense

The Mythos deployment raises fundamental questions about AI governance in defense. Can a government agency deploy technology from a blacklisted vendor? The current situation suggests the answer, at least temporarily, is yes. Officials justify this through the lens of urgent national security need.

The Project Glasswing framework offers a possible model for future deployments. It separates defensive use from broader commercial relationships. It allows governments to access cutting-edge tools without full vendor endorsement. This kind of structured access program may become increasingly important as AI capabilities advance.

Emil Michael‘s statements highlight the speed at which this field moves. Decisions that once took months now require responses in days. The government’s ability to adapt quickly will determine its effectiveness. Mythos represents one attempt to stay ahead of that accelerating curve.

What makes this story significant is its broader signal to the industry. Governments will use the most capable tools available, regardless of political friction. The race to build the most powerful defensive AI has real national security stakes. Every major AI company now operates with that reality clearly in view.