The US National Security Agency is reportedly using Anthropic’s advanced AI model even as the Pentagon continues to flag the company as a risk.
Quick Summary – TLDR:
- NSA is using Anthropic’s Mythos AI model despite Pentagon restrictions.
- Pentagon labelled Anthropic a supply chain risk earlier this year.
- AI model focuses on cybersecurity, including vulnerability detection.
- Disagreement stems from limits on AI use, especially surveillance and weapons.
What Happened?
The National Security Agency has reportedly adopted Anthropic’s Mythos Preview AI model, even though the Department of Defense has raised concerns about the company’s reliability. The situation highlights a clear divide within the US government on how advanced AI tools should be used in national security.
While defense officials have tried to limit engagement with Anthropic, intelligence agencies appear to be moving forward with deployment to meet operational needs.
JUST IN: The NSA has reportedly been using Claude Mythos despite Anthropic’s “supply chain risk” designation.
— Polymarket (@Polymarket) April 19, 2026
A Government Divided on AI Adoption
At the center of the issue is Anthropic’s Mythos Preview, the company’s most advanced AI model to date. The model is designed with strong cybersecurity capabilities, including the ability to identify software vulnerabilities and prevent cyberattacks.
Despite its capabilities, the Department of Defense labelled Anthropic a “supply chain risk” and reportedly instructed vendors earlier this year to restrict its use. This move came after disagreements over how the AI should be deployed across defense systems.
However, sources indicate that the NSA has continued to use Mythos, with some usage extending beyond a single unit. This creates a contradictory situation where one part of the government is restricting the technology while another is actively using it.
Why the Pentagon Raised Concerns?
The tension between Anthropic and the Pentagon began during contract renegotiations earlier this year. Defense officials reportedly pushed for broader access to Anthropic’s AI models for “all lawful purposes.”
Anthropic resisted certain demands, especially those related to:
- Mass domestic surveillance
- Autonomous weapons systems
This stance has caused friction. Some Pentagon officials believe these restrictions could limit the AI’s usefulness in defense scenarios, raising concerns about whether the company can fully support military requirements.
Anthropic, on the other hand, has maintained its position, drawing clear boundaries on how its AI should be used.
Mythos AI Gains Ground Despite Limits
Even with these disagreements, Mythos is gaining traction inside government and industry. Anthropic has tightly controlled access to the model, limiting it to around 40 organizations globally due to concerns over misuse, especially in offensive cyber operations.
Among these organizations are major tech companies such as:
These firms are using the model to scan systems, detect weaknesses, and strengthen security infrastructure before attackers can exploit vulnerabilities.
Although exact details of NSA usage remain unclear, experts believe the agency is likely using the model for similar defensive cybersecurity purposes.
Global Interest and Expanding Reach
The reach of Mythos may not be limited to the United States. Reports suggest that partners in the United Kingdom may also have access through national AI security programs, indicating growing global demand for such advanced cybersecurity tools.
Meanwhile, discussions between Anthropic leadership and US officials appear to be ongoing. CEO Dario Amodei recently met with senior figures including Susie Wiles and Scott Bessent.
These talks reportedly focused on AI deployment strategies and national security concerns, and sources describe them as productive. The outcome could influence how more US agencies adopt AI tools going forward.
SQ Magazine’s Takeaway
I think this situation clearly shows how AI adoption is moving faster than policy can keep up. On one side, the Pentagon is worried about control and long term risks. On the other, agencies like the NSA cannot afford to ignore tools that can actively prevent cyber threats today.
In my view, this is not just about Anthropic. It is about a bigger shift where governments are being forced to choose between strict control and immediate capability. And right now, capability seems to be winning.
