Anthropic says its AI spotted vulnerabilities in high-security government systems

AI model reportedly found weaknesses in classified systems

What happens when an AI system is pointed at some of America’s most secure computer networks? According to a U.S. official, one of Anthropic’s advanced AI models quickly spotted vulnerabilities in classified government systems during a recent testing exercise.

The claim immediately raised eyebrows because the model reportedly identified the weaknesses in just hours. While officials said the AI did not necessarily exploit those vulnerabilities, the results highlighted how powerful advanced AI tools are becoming.

The tests involved Anthropic’s highly restricted Mythos model

The testing reportedly used Anthropic’s Mythos model, one of the company’s most advanced AI systems. Access to the model has been tightly limited because of concerns about its cybersecurity capabilities.

Officials said the model identified certain weaknesses in secure government systems during the exercise. The findings have added to growing debates about how powerful AI systems should be handled and monitored.

A senator revealed the surprising test during a hearing

The testing first gained public attention during a Senate hearing earlier this month. Sen. Mark Warner of Virginia said the tool had broken into almost all classified systems in hours instead of weeks.

His remarks suggested the AI’s capabilities were significant enough to concern national security officials. The comments also sparked questions about how prepared governments are for increasingly capable AI systems.

The effort was part of a project called Glasswing

The testing was reportedly conducted through Project Glasswing, an Anthropic initiative focused on helping selected partners find and fix vulnerabilities in important software systems.

AP reported that Anthropic also worked with U.S. intelligence agencies on the classified-system testing. The effort reflects growing cooperation between AI companies and government agencies as officials assess how advanced AI could affect cybersecurity and national security.

Anthropic and the Trump administration have clashed

Despite cooperating on security testing, Anthropic and the Trump administration have recently disagreed over how advanced AI systems should be handled. The company has raised concerns about military uses of its technology.

At the same time, the administration has imposed restrictions on some of Anthropic’s models. The disagreements show how difficult it has become to balance innovation, national security, and public safety concerns.

Little-known fact: The average global cost of a data breach fell to $4.4 million in 2025, down 9 percent from the previous year, largely because organizations got faster at identifying and containing cyber incidents.

Finding a flaw is not the same as exploiting it

Officials stressed an important detail about the exercise. Although the AI reportedly found vulnerabilities within hours, that does not mean it was able to break into or exploit those systems in the same amount of time.

Security experts often distinguish between identifying weaknesses and successfully using them to gain access. Even so, quickly finding vulnerabilities could dramatically speed up cybersecurity work.

The government ordered new restrictions on the company’s models

The administration issued a June 12 directive requiring Anthropic to prevent foreign nationals from using Fable 5 and Mythos 5. Anthropic said it disabled both models for all customers to comply, while arguing that the government’s response was unwarranted.

The situation has since shifted. Reuters reported on June 26 that the U.S. government allowed Anthropic to redeploy Claude Mythos 5 to some trusted U.S. organizations, while Anthropic said it was still working to expand access to Mythos 5 and make Fable 5 generally available again.

A new executive order puts advanced AI under review

President Donald Trump recently signed an executive order creating a framework for reviewing the national security risks of advanced AI systems before they are publicly released.

The order allows the federal government to examine the risks posed by leading AI systems for up to a month. Participation by AI developers, however, remains voluntary under the framework.

Cybersecurity experts pushed back against the restrictions

More than 100 cybersecurity leaders and experts have reportedly asked the administration to lift its restrictions on Anthropic’s models. They argued that limiting access could hurt America’s defenses.

The group said advanced AI tools are useful for identifying software flaws and training security teams. Taking away strong defensive tools, they argued, could benefit adversaries that continue advancing their own capabilities.

The debate is becoming bigger than one AI company

The Anthropic testing exercise highlights a growing reality. Advanced AI systems are becoming powerful enough to transform both cybersecurity and national security planning.

Governments now face difficult choices about how to regulate these tools while still benefiting from their defensive capabilities. The discussion is likely to intensify as AI systems continue improving.

AI could become both a cyber shield and a cyber threat

The same AI systems that help uncover vulnerabilities could also create new risks if they fall into the wrong hands. That possibility is driving many of the current policy debates around advanced models.

As governments and companies race to understand these technologies, AI is increasingly being viewed as both a powerful security tool and a potential cybersecurity challenge.

If you’re curious about what happens when cutting-edge AI leaves the lab and reaches the public, take a look at how Anthropic unveils civilian Mythos amid safety debate.

A glimpse at how AI could reshape national security

The report that Anthropic’s AI found vulnerabilities in classified systems offers a glimpse into how quickly artificial intelligence is changing cybersecurity. Even limited testing results have sparked major questions about preparedness and oversight.

The findings suggest advanced AI could become an important tool for defending critical systems while also creating new challenges.

If you’re curious about how concerns over AI’s real-world impact are expanding, take a look at why Florida launches an inquiry into OpenAI following the FSU shooting.

What do you think about AI’s growing role in national security? Share your thoughts.

This slideshow was made with AI assistance and human editing.

Don’t forget to follow us for more exclusive content on MSN.

Read More From This Brand:

• Microsoft experiments with Copilot AI that can operate across your desktop
• Microsoft faces scrutiny over alleged Chinese engineers handling DOD systems
• Meta reportedly gains two more elite AI researchers from OpenAI