Main Points of this Article
- U.S. NSA introduces advanced AI “Mythos”, contradicting the Pentagon’s elimination policy
- White House prepares to provide to government agencies, revamps cyber defenses
U.S. NSA introduces advanced AI model “Mythos”
The U.S. National Security Agency (NSA) has introduced “Claude Mythos Preview”, Anthropic‘s advanced AI model into a confidential network, according to a report by Axios on the April 20, 2026.
This situation can be observed as unusual because the Department of Defense (DoD), which supervises the NSA, has tagged Anthropic as a “supply chain risk” and is currently battling it out in the courts to remove the company’s products.
Issue Backgrounder
The Pentagon and Anthropic were initially cooperative. They signed a contract worth up to $200 million in July 2025, making Claude the first AI model approved in a network environment that is considered sensitive, even if the agreement included the company’s safety policy of prohibiting its usage in “domestic large-scale surveillance” and “fully autonomous weapon systems.”
Conflict between the two became definitive only this 2026. While the Pentagon demanded the rescinding of the clause for unrestricted use “for all lawful purposes,” Anthropic refused, citing safety, and after a final deadline on February 27, the Pentagon took an initiative to effectively remove the company as a supply chain risk.
Cooperation in the White House
In the midst of the legal battle, the White House Office of Budget Management (OMB) notified key departments, such as the Treasury Department, State Department, and Department of Justice, on April 17 that they had begun preparing equipment for the safe use of Mythos. Also, on the same day, it was reported that Trump administration chief of staff Wiley and Treasury Secretary Bessent had a meeting with Dario Amodey, Anthropic CEO, and had “productive discussions” with regard to the administration’s introduction of technology.
Moreover, Treasury Secretary Bessent and Federal Reserve Chairman Powell collaborated to recommend security testing using Mythos to major banks. The momentum to focus on practical benefits is also accelerating. In public statements, Anthropic is concerned about the high attack ability of Mythos to identify vulnerabilities in operating systems and browsers. They continue to cooperate with the government to build appropriate safeguards.
On the other hand, on April 15, the security research team revealed the results of a survey that the unique vulnerability-finding capabilities of Mythos can be replicated in publicly available AI models such as GPT-5.4. With the availability of AI with such capabilities in public environments, the “risk of model leakage” exploited by individuals and by government hackers has become a new threat to the entire digital infrastructure, including cryptocurrencies and blockchains.
The introduction of this by the NSA, which continues have unresolved conflicts within the U.S. government, reflects the current situation where acquiring top-of-the-line AI has become a top national security priority. In response to the change in mindset in defense and attack brought about by artificial intelligence, such as identifying weaknesses in cryptocurrency smart contracts, U.S. government agencies are still divided regarding their adoption and risk management.
