The US National Security Agency is quietly testing Anthropic’s Mythos AI to detect software vulnerabilities, highlighting its growing role in cybersecurity. Early feedback points to strong performance, even as political tensions and security concerns in Washington continue to shape how such powerful AI systems are deployed and controlled.

Anthropic’s latest artificial intelligence system, Mythos, is drawing attention at the highest levels of the US government. According to a Bloomberg report, the US National Security Agency (NSA) has been actively testing the model’s ability to identify vulnerabilities in widely used software, including products developed by Microsoft.

The evaluation comes at a time when advanced AI tools are increasingly seen as both a strategic asset and a potential risk, particularly in cybersecurity and national defence contexts.

STORY CONTINUES BELOW THIS ADEarly tests point to speed and precision

People familiar with the matter told Bloomberg that NSA officials have been impressed by Mythos’s performance. The system is said to rapidly scan complex codebases and flag possible weaknesses, potentially accelerating a process that traditionally requires extensive manual effort.

Teams within the agency, including its cybersecurity directorate, are comparing Mythos’s outputs with existing internal tools and research methods. However, it remains unclear whether the testing has uncovered any previously unknown vulnerabilities.

Anthropic has deliberately limited access to Mythos, offering it only to select organisations. The company has encouraged these early users, which include major technology firms and financial institutions, to deploy the tool defensively, identifying weaknesses before malicious actors can exploit them.

Executives ultimately decided against a broader public release, citing the risk that the same capabilities could be weaponised by hackers. We previously reported that the
White House has also pushed back against plans to expand access to dozens more organisations.

Policy friction complicates adoption

The NSA’s evaluation is unfolding against a backdrop of political tension. The Pentagon has labelled Anthropic and its technology a potential supply chain risk, with US Defence Secretary Pete Hegseth earlier signalling a move away from the company’s services.

Anthropic has challenged that designation in court and recently secured a ruling that temporarily blocks restrictions on government use of its technology. The company argues that such limits could significantly impact its business prospects.

Anthropic CEO Dario Amodei has reportedly engaged with White House officials to explore ways the government might safely use versions of Mythos. As agencies like the NSA continue testing, the outcome could influence how AI-driven cybersecurity tools are adopted more broadly.

For now, Mythos sits at the intersection of innovation and caution, offering powerful capabilities while raising equally significant questions about control, access and unintended consequences.

STORY CONTINUES BELOW THIS AD

First Published:
May 01, 2026, 08:36 IST

HomeTechUS security agency quietly tests Anthropic Mythos AI to hunt software vulnerabilitiesEnd of Article