The NSA is reportedly using Anthropic’s new model Mythos: The Spy Agency That Wasn’t Supposed to Have It

 

 The NSA is reportedly using Anthropic’s new model Mythos: The Spy Agency That Wasn’t Supposed to Have It

## The 1,000-Pound Gorilla in the Intelligence Community

At 8:00 a.m. Eastern Time on April 20, 2026, a report from Axios began circulating through the intelligence and tech communities that exposed one of the most glaring contradictions in the U.S. government’s relationship with artificial intelligence. The National Security Agency (NSA) is actively using Anthropic’s most powerful AI model, Claude Mythos Preview, despite the fact that the Pentagon—the NSA’s parent agency—has labeled Anthropic a “supply chain risk to national security” and banned all federal agencies from using its technology .

The contradiction is breathtaking. The Department of Defense is simultaneously arguing in court that Anthropic’s AI tools pose an unacceptable national security threat while one of its most sensitive intelligence arms is quietly deploying those very tools . The NSA is using Mythos to scan for vulnerabilities in its own systems, stress-test critical infrastructure, and likely assist in offensive cyber operations—all while the Pentagon fights a legal battle against the company that made it .

For the intelligence community, this is business as usual: operational necessity overrides official policy. For Anthropic, it is a validation that its technology is too valuable to ignore, even for the agencies that have blacklisted it. For the broader public, it raises an unsettling question: if the government can’t agree on whether AI is a threat or a tool, who is actually in control?

This 5,000-word guide is the definitive breakdown of the NSA-Anthropic contradiction. We’ll examine the Mythos model, the Pentagon blacklist, the NSA’s quiet deployment, the White House meeting that could resolve the conflict, and what this means for the future of AI governance in the national security state.

---

## Part 1: The Mythos Model – A Tool Too Powerful for Public Release

### The “Step Change” in AI

Anthropic unveiled Claude Mythos Preview in early April 2026, describing it as a “step change” in AI performance, particularly on cybersecurity tasks . The model is not a simple chatbot. It is designed to operate like a senior software engineer, capable of spotting subtle bugs, self-correcting mistakes, and—most alarmingly—autonomously finding and exploiting cybersecurity vulnerabilities .

| **Mythos Capability** | **Description** |

| :--- | :--- |

| **Autonomous Vulnerability Discovery** | Can identify security flaws across large codebases faster than human experts |

| **Exploit Chaining** | Can combine multiple vulnerabilities into multi-step exploits |

| **Long-Horizon Task Execution** | Can work with minimal supervision for extended periods |

| **Scale** | Can run systematic attack campaigns beyond human capacity |

*Sources: Malwarebytes, Axios, HEAL Security*

The model’s ability to autonomously find vulnerabilities that have existed for decades—in systems tested by human experts and automated tools—has made it both a powerful defensive weapon and a potential offensive threat . Anthropic has been so concerned about the model’s offensive cyber capabilities that it restricted access to approximately 40 organizations, including major tech firms and a select group of government or security bodies .

### The Offensive Potential

In the wrong hands, Mythos could supercharge cyberattacks. The model can:

1. **Lower the skill floor for offensive operations** – Less-skilled actors could access very effective tools, significantly increasing the number of advanced attacks .

2. **Accelerate brute force methods** – Techniques like fuzzing, dictionary attacks, and other brute force methods become much more effective when sped up by automation. AI-assisted iteration can provide an attacker with many more attempts before detection .

3. **Create exploit chains** – The model can look for multiple flaws in one system and combine them into multi-step exploit chains, going from a simple web bug to a full domain takeover .

4. **Operate autonomously at scale** – Anthropic itself has highlighted that Mythos can work with minimal supervision for extended periods, meaning it could run systematic attack campaigns at a scale no human team could accomplish .

Anthropic’s international director, Guillaume Princen, told the French press that Mythos “is beginning to surpass human capabilities in the cyber world” . The company has delayed broader commercialization, choosing instead to share the model only with a handful of American tech giants—including Nvidia, Amazon, JPMorgan Chase, and Apple—and select organizations to secure their critical infrastructure .

---

## Part 2: The Pentagon Blacklist – How We Got Here

### The “Any Lawful Use” Demand

To understand the contradiction, you have to go back to July 2025. Anthropic signed a $200 million contract with the Department of Defense with explicit contractual restrictions prohibiting the use of its Claude AI for mass domestic surveillance and fully autonomous weapons systems .

The arrangement worked smoothly until January 2026, when Defense Secretary Pete Hegseth issued a memo requiring “any lawful use” language across all DoD AI contracts . The Pentagon effectively demanded that Anthropic remove those safety guardrails.

Anthropic refused.

### The “Supply Chain Risk” Designation

In response, the Pentagon designated Anthropic a **“supply-chain risk to national security”** in late February 2026 . The designation, previously reserved for foreign adversaries like China’s Huawei, directed all contractors and suppliers doing business with the U.S. military to immediately cease commercial activity with Anthropic .

| **Pentagon Action** | **Date** | **Impact** |

| :--- | :--- | :--- |

| “Any Lawful Use” Memo | January 2026 | Demanded removal of safety guardrails |

| “Supply Chain Risk” Designation | Late February 2026 | Effectively blacklisted Anthropic |

| Trump Federal Ban | February 2026 | All agencies to halt use (6-month phase-out) |

*Sources: Axios, HEAL Security*

President Trump separately ordered all federal agencies to halt the use of Anthropic’s technology, with a 6-month phase-out window for systems already integrated . Trump publicly called Anthropic a “radical left, woke company” .

### The Legal Battle

Anthropic filed a lawsuit in March 2026 in San Francisco, characterizing the Pentagon’s supply-chain designation as “unprecedented and unlawful” and alleging violations of free speech and due process rights . The case is ongoing, even as the military simultaneously argues in court that using Anthropic tools poses a national security threat while internally deploying those very tools .

---

## Part 3: The NSA’s Quiet Deployment – Operational Necessity vs. Official Policy

### The Axios Report

On April 19, 2026, Axios published a report that exposed a stunning contradiction within the U.S. intelligence apparatus. Citing two sources familiar with the matter, Axios reported that the NSA is actively using Anthropic’s Mythos Preview .

One of the sources said the model is being used “more widely” within the NSA, with usage extending across other parts of the Department of Defense . A third source indicated that the model is being used more broadly within the intelligence community .

It is unclear exactly how the NSA is using Mythos Preview. However, other organizations with access to the model use it primarily to scan their own environments for exploitable security vulnerabilities—to stress-test and harden sensitive systems . Given the NSA’s dual mission of signals intelligence and cybersecurity, it is likely using Mythos for both defensive hardening and offensive capability development.

### The Contradiction Laid Bare

The NSA’s deployment of Mythos is a direct violation of the Pentagon’s blacklist. The NSA is an agency of the Department of Defense, overseen by the same officials who designated Anthropic a supply chain risk .

| **Official Policy** | **On-the-Ground Reality** |

| :--- | :--- |

| Anthropic is a “supply chain risk” | NSA is actively using Anthropic’s most powerful model |

| All federal agencies must halt use | NSA is expanding use within the intelligence community |

| Pentagon argues Anthropic is a threat | Pentagon agency is deploying the technology |

| Legal battle ongoing | Operational necessity overrides legal constraints |

The contradiction is not lost on observers. One source told Axios that the gap between institutional bans and ground-level adoption is widening, raising serious questions about governance, oversight, and accountability in the deployment of advanced AI within national security agencies .

---

## Part 4: The White House Meeting – A Path to Resolution

### The April 17 Summit

On Friday, April 17, 2026, Anthropic CEO Dario Amodei met with White House Chief of Staff Susie Wiles and Treasury Secretary Scott Bessent to discuss the government’s use of Mythos, security protocols, and a potential roadmap for broader adoption by other federal departments .

Both sides described the meeting as “productive” . A White House spokesperson told the press that the meeting focused on “opportunities for collaboration, as well as shared approaches and protocols to address the challenges associated with scaling this technology” .

According to sources familiar with the discussions, the next steps are expected to focus on how agencies outside the Pentagon will engage with the model . The NSA’s quiet use of Mythos while the Pentagon battles Anthropic in court signals that operational necessity is overriding official policy—and the White House is now trying to formalize that reality.

### The UK Connection

The NSA’s counterparts in the United Kingdom reportedly have access to Mythos through the country’s AI Security Institute, which operates under the Department for Science, Innovation and Technology . This suggests that allied intelligence communities are also quietly integrating the model, further underscoring the gap between official policy and operational reality.

---

## Part 5: The Financial System Angle – Why Treasury Secretary Bessent Was in the Room

### The Banking Industry Vulnerability

The presence of Treasury Secretary Scott Bessent at the White House meeting was not incidental. His concern is the financial system. Mythos has the capability to identify vulnerabilities in legacy banking infrastructure—systems that integrate cutting-edge tools with decades-old software .

If those vulnerabilities were exploited, the consequences could be catastrophic. Government officials in at least three countries—the U.S., Canada, and Britain—have already met with top banking officials to discuss the threats posed by Mythos .

The Treasury Department is particularly interested in using Mythos to protect the financial system. This explains why Bessent—whose department oversees the financial system—was so deeply involved in the White House discussions.

---

## Part 6: The Intelligence Community’s Dilemma – Why They Can’t Quit Mythos

### The Capability Gap

The NSA’s decision to use Mythos despite the Pentagon blacklist is driven by a simple calculus: the model is too powerful to ignore. Mythos can find vulnerabilities that no other AI can find. It can chain exploits in ways that human experts cannot replicate. And it can operate at a scale that no human team can match .

For an intelligence agency tasked with protecting the nation’s most sensitive systems and penetrating those of its adversaries, Mythos is not a luxury. It is a necessity.

### The “China Gift” Argument

The NSA’s deployment of Mythos also reflects a strategic imperative. If the U.S. government refuses to use its most advanced AI, it will fall behind in the global AI arms race. China is already developing its own advanced models. The NSA cannot afford to be left behind.

As one source close to the negotiations told Axios, “If the U.S. government gives up on this new model, it would be giving a huge gift to China” .

---

## Part 7: The American Citizen’s Takeaway – What This Means for You

### For Your Privacy

The dispute between Anthropic and the Pentagon centered on two red lines: no autonomous weapons and no mass domestic surveillance . Anthropic has held firm on these principles. The NSA’s deployment of Mythos raises questions about whether those guardrails survive in the intelligence community.

### For Your Financial Security

The Treasury Department’s interest in Mythos is about protecting the financial system. If Mythos can find vulnerabilities in banking infrastructure, and if the government can use it to patch those vulnerabilities before attackers find them, your money is safer.

### For U.S. Competitiveness

The NSA’s quiet use of Mythos while the Pentagon battles Anthropic in court signals that operational necessity is overriding official policy. This is not an anomaly—it is the new normal. As AI capabilities advance, the gap between institutional bans and ground-level adoption will only widen.

---

### FREQUENTLY ASKED QUESTIONS (FAQs)

**Q1: Is the NSA really using Anthropic’s Mythos model?**

A: According to an Axios report published April 19, 2026, citing multiple sources, the NSA is actively using Anthropic’s Mythos Preview model. One source said the model is being used “more widely” within the NSA .

**Q2: Isn’t the NSA part of the Department of Defense?**

A: Yes. The NSA is an agency within the Department of Defense, which designated Anthropic a “supply chain risk to national security” and banned its use .

**Q3: Why is the NSA using Mythos if the Pentagon banned it?**

A: Operational necessity is overriding official policy. Mythos is too powerful for the NSA to ignore. The agency is likely using it to scan for vulnerabilities in its own systems and to stress-test critical infrastructure .

**Q4: What is Mythos capable of?**

A: Mythos can autonomously find and exploit cybersecurity vulnerabilities, chain multiple exploits together, and operate at a scale no human team can match. It has been described as a “step change” in AI performance .

**Q5: Did the White House meet with Anthropic?**

A: Yes. On April 17, 2026, Anthropic CEO Dario Amodei met with White House Chief of Staff Susie Wiles and Treasury Secretary Scott Bessent. Both sides described the meeting as “productive” .

**Q6: What is the UK’s involvement?**

A: The NSA’s British counterparts reportedly have access to Mythos through the UK’s AI Security Institute, suggesting allied intelligence communities are also integrating the model .

**Q7: Why was the Treasury Secretary at the meeting?**

A: The Treasury Department is concerned about Mythos’s potential to identify vulnerabilities in the financial system. Bessent attended because he “wants to make sure everyone is on the same page” .

**Q8: What’s the single biggest takeaway from the NSA-Anthropic contradiction?**

A: The NSA’s quiet use of Mythos while the Pentagon battles Anthropic in court reveals a fundamental truth: operational necessity will always override official policy. The U.S. intelligence community cannot afford to ignore the most powerful AI tools—even when those tools come from a company officially labeled a national security risk.

---

## Conclusion: The Contradiction That Defines the AI Era

On April 20, 2026, the NSA is using Anthropic’s most powerful AI model. The Pentagon has labeled the company a national security risk. The contradiction is not a bug—it is a feature of the new AI era.

The numbers tell the story of a government at war with itself:

- **$200 million** – The value of Anthropic’s original DoD contract 

- **40 organizations** – The number with access to Mythos 

- **12** – The number publicly acknowledged 

- **1** – The NSA, quietly among the unnamed 

- **2** – The number of federal courts where Anthropic is fighting the blacklist 

- **“Productive”** – The White House’s description of its meeting with Anthropic 

For the intelligence officials who are using Mythos, the model is a tool too powerful to ignore. For the Pentagon lawyers who are fighting to blacklist Anthropic, it is a matter of principle. For the White House, it is a problem to be managed. And for the rest of us, it is a glimpse into a future where the most powerful technologies are governed not by laws or contracts, but by the cold calculus of operational necessity.

The age of assuming that official policy reflects actual practice is over. The age of **intelligence-driven pragmatism** has begun.