While Dario Amodei publicly debates the safety of humanity, the U.S. National Security Agency (NSA) is actively deploying Anthropic’s Mythos model to dismantle infrastructure in China and Iran. According to the Financial Times, this is not a case of passive monitoring, but full-scale participation in offensive operations. To tailor these algorithms for intelligence needs, Anthropic has embedded a team of six engineers directly within NSA headquarters. These specialists are now fine-tuning Mythos to exploit vulnerabilities and breach secure networks, effectively turning yesterday’s "safe AI" into a digital master key.
The situation is particularly striking given Anthropic’s long-standing friction with the Pentagon. Previously, the U.S. Department of Defense even placed the company on a supply chain risk list, attempting to block it from government contracts. This tension stemmed from Anthropic’s refusal to allow its Claude model to be used in killer drones or mass surveillance systems. However, these ethical prohibitions apparently do not apply to Mythos. Evidently, in Amodei’s moral calculus, offensive cyber-hacking is more ethical than missile guidance—or perhaps the company simply received an offer it couldn't refuse.
Double Standards and Global Expansion
While carrying out classified assignments for the intelligence community, Anthropic remains focused on its bottom line. Access to the Mythos model has already been granted to 150 organizations across 15 countries. This strategy of double standards—arming intelligence agencies while simultaneously scaling in the global commercial market—is unfolding against the backdrop of new regulatory initiatives from the Trump administration. By signing onto voluntary safety testing pledges, Anthropic is attempting to maintain its image as a high-minded vendor, even as its engineers sharpen the tools of digital warfare on the front lines.
Anthropic’s transformation from an AI safety evangelist into a full-scale participant in the arms race is a game-changer. When the lines between defensive alignment and offensive efficiency blur, it becomes clear: in the world of realpolitik, "safety" is merely a marketing label for a tool designed to strike first.
Key Takeaways from Anthropic’s Strategic Pivot
The Mythos model is being utilized by the NSA for offensive cyber operations against sovereign states. Anthropic engineers are embedded at intelligence headquarters to develop vulnerability exploitation tools. The company is balancing classified military contracts with commercial expansion across 15 countries. The ethics of "safe AI" have taken a backseat to U.S. national security interests.
If a model is optimized to find holes in foreign networks, any talk of ethics becomes a sophisticated smokescreen for serving the cyber-front.