Lina Irawan
In an unexpected twist that highlights the complex, often contradictory landscape of modern geopolitics and artificial intelligence, reports have emerged suggesting that the National Security Agency (NSA) is actively utilizing Anthropic’s cybersecurity-focused AI, "Mythos," for offensive cyber operations. This revelation arrives only months after the U.S. Department of Defense (DOD) formally designated Anthropic as a "supply chain risk," effectively barring the firm from government contracts.
The report, first surfaced by the Financial Times, suggests that despite the high-profile public fallout between the AI lab and the defense establishment, a quiet, tactical collaboration persists. With roughly half a dozen Anthropic engineers reportedly embedded within the NSA, the arrangement signals that the strategic necessity of maintaining a technological edge in the global cyber arms race may be superseding political disputes.
A Chronology of Conflict: From Partnership to Blacklist
The current friction between Anthropic and the U.S. government is rooted in the events of early 2026. At the time, the two parties were engaged in advanced negotiations for a $200 million contract intended to integrate Anthropic’s cutting-edge AI capabilities into defense infrastructure.
However, the talks collapsed under the weight of ideological differences. The Trump administration, as a condition of the partnership, demanded that Anthropic ensure its models were available for "all lawful purposes." This directive, interpreted by many in the industry as a mandate to disable critical safety guardrails, placed Anthropic in an untenable position. CEO Dario Amodei publicly resisted, arguing that his firm’s technology should not be leveraged for activities like mass surveillance or the development of autonomous weapon systems—tools that, in his view, could fundamentally undermine democratic values.
The fallout was swift. The DOD terminated the contract negotiations, and in March 2026, the administration took the unprecedented step of labeling the San Francisco-based Anthropic a "supply chain risk." By categorizing the company alongside entities like Huawei and ZTE, the government effectively blacklisted it from working with the DOD. Anthropic responded by filing a lawsuit, claiming the designation was a violation of the First Amendment, setting the stage for a protracted legal battle that currently remains unresolved.
The Mythos Mandate: Why the NSA is Doubling Down
The reported utilization of "Mythos" within the NSA provides a stark contrast to the DOD’s official stance. Mythos is an specialized AI model designed specifically for identifying and mitigating cybersecurity vulnerabilities.
According to sources familiar with the arrangement, the logic driving this partnership is simple: the U.S. cannot afford to lag behind. As the time required to exploit a "zero-day" vulnerability shrinks—moving from months to, in some projections, mere minutes—the ability to automate cyber defense and offense has become a cornerstone of national security.
The NSA’s interest in Mythos is specifically focused on its potential for offensive cyber operations, including the infiltration of networks belonging to adversarial nations such as China and Iran. While it remains unclear whether the embedded Anthropic staff are assisting in active, real-time cyber warfare or are primarily focused on the customization and technical optimization of the model, their presence within the agency confirms that Mythos is considered an indispensable asset by the intelligence community.
Navigating the Contradictions: The OpenAI Factor
The complexity of the situation is further compounded by the DOD’s parallel engagement with Anthropic’s primary competitor, OpenAI. Following the blacklisting of Anthropic, the Pentagon moved to formalize a partnership with OpenAI.
This move raised significant questions regarding the consistency of the administration’s stance. While the DOD has claimed that its procurement decisions are based on security and compliance, critics point out that OpenAI, like Anthropic, has been vocal about the need for AI safety. When the DOD accepted certain safety-related clauses from OpenAI in their subsequent deal, it reinforced the perception that the initial rift with Anthropic may have been driven more by political friction than by genuine security concerns.
The legal landscape is equally murky. Two separate courts have issued contradictory rulings regarding the DOD’s labeling of Anthropic as a risk. These conflicting judicial perspectives ensure that, for the moment, the legal status of AI vendors within the federal government remains in a state of flux.
The Executive Order and the New Regulatory Reality
This intelligence-sector reliance on Mythos arrives during a period of heightened federal oversight. A recent executive order signed by the Trump administration mandates that AI companies provide the government with access to their "frontier models" for a 30-day review period prior to public release.
While the order is intended to prevent the proliferation of dangerous capabilities, it has created a new, ambiguous category of "trusted partners." The ongoing, albeit quiet, collaboration between the NSA and Anthropic suggests that even companies currently at odds with the executive branch may find themselves pulled back into the government’s orbit due to the sheer utility of their products. The lack of clarity regarding what defines a "covered frontier model" leaves the industry bracing for further regulatory surprises.
Implications for AI Development and Global Security
The implications of this situation are twofold: they speak to the immense power of current AI models and the fragility of private-sector independence in the face of national security imperatives.
1. The Proliferation of Offensive AI
As AI models become more capable of identifying, analyzing, and exploiting software vulnerabilities, the "security" of the global internet is becoming increasingly tied to the intelligence community’s control of these tools. If the NSA successfully integrates Mythos into its offensive operations, it sets a precedent that will almost certainly be mimicked by every major global power. This accelerates an arms race where AI is not just a tool for commerce, but the primary weapon for geopolitical maneuvering.
2. The Erosion of Corporate Agency
Anthropic’s situation demonstrates how difficult it is for AI companies to maintain their "safety-first" ethos when the state comes calling. Even while in the midst of a lawsuit against the DOD, the firm is seeing its technology utilized by a core component of the U.S. intelligence apparatus. This suggests that in the era of AI, the distinction between private enterprise and state security is rapidly dissolving.
3. A Shift in Technical Development
Perhaps most tellingly, Anthropic recently revealed that its Claude model is now contributing to its own codebase, writing over 80% of its merged code. This rapid self-improvement cycle—which has led Anthropic to publicly call for the option to slow down development—is exactly what makes these models so attractive to the NSA. The ability for an AI to evolve faster than human programmers can track makes it an asset of immeasurable value for cyber warfare.
Looking Ahead: A Fragile Future
As Mythos moves from early access to broader distribution—currently planned for 150 organizations across 15 countries—the question remains: can Anthropic maintain its principles while its technology is simultaneously fighting a legal battle against the state and being used by that same state to conduct cyber operations?
The NSA’s reported use of Mythos suggests that the U.S. government is willing to overlook its own formal designations when the alternative is losing a strategic advantage. For the engineers embedded in the agency and the executives at Anthropic, the situation is a delicate balancing act. They are operating in the shadow of a public dispute, performing critical work for the national security apparatus while the legal battle over their right to exist as a government contractor continues in the courts.
Ultimately, the Mythos case serves as a microcosm of the 21st-century technological landscape: a world where safety guardrails, corporate values, and legal standing are constantly being renegotiated against the relentless, high-stakes demand for national security and digital dominance. Whether this clandestine arrangement leads to a reconciliation between the DOD and Anthropic, or further fuels the fires of the ongoing conflict, will be a defining question for the trajectory of AI in the years to come.
Nana Wu
Siti Muinah
