NSA Loses Access to Anthropic's Mythos AI Tool

NSA Access to Mythos Terminated Amid Contract Dispute

The National Security Agency (NSA) has lost access to Mythos, an AI tool developed by Anthropic, due to a dispute over a classified contract. The White House and intelligence officials had previously attempted to establish a formal agreement that would allow the NSA to utilize Anthropic's technology for intelligence analysis and the detection of new computer vulnerabilities.

Capabilities and Security Implications of Mythos

Mythos is characterized as a highly capable tool for cybersecurity and intelligence. Reports indicate that the model's efficiency in penetrating secure systems is significant, with one account stating that Mythos "broke into almost all of our classified systems, not in weeks, but in hours."

Technical discussions suggest that the tool's power lies in its ability to process vast amounts of data, such as decompilation output from tools like IDA Pro, to bypass "security by obscurity." By analyzing code at scale, the LLM can identify vulnerabilities that would otherwise remain hidden from human analysts.

Government and Corporate Tension

The dispute reflects a broader conflict regarding the control and classification of advanced AI models. The U.S. government views these models as critical national security assets, while private companies like Anthropic navigate the balance between safety, commercial release, and government cooperation.

Divergent Perspectives on Model Control

There is a perceived contradiction in how both the government and AI labs frame these technologies:

• Government Position: The government views the model as a supply chain risk while simultaneously attempting to use the Defense Production Act (DPA) to compel access.
• Anthropic Position: The company frames the model as having national security implications akin to a nuclear weapon, yet maintains that it is safe for general public release.

Community Skepticism and Counterpoints

Technical observers and commenters have expressed skepticism regarding the total loss of access, questioning whether a government agency as powerful as the NSA could be completely locked out of a critical technology.

Doubts Regarding Access Loss

Some argue that the NSA may already possess the model weights or have internal plants within AI companies, suggesting the reported "loss of access" may be a PR maneuver or a result of administrative incompetence rather than a technical lockout.

Concerns Over AI-Driven Vulnerabilities

Beyond the specific dispute, there is growing concern that the speed at which AI models can find vulnerabilities is far outpacing the speed at which humans can patch them. As multiple models catch up in capability, the window for securing underlying infrastructure is shrinking.

Critique of AI Marketing

Some critics view the narrative surrounding Mythos's "danger" as a marketing stunt designed to increase the perceived value and prestige of Anthropic by associating its tools with the NSA's requirements.