Days after Anthropic unveiled Claude Mythos, U.K. Minister Kanishka Narayan has claimed that the frontier AI model has solved their “hardest cyber end-to-end” test, as he stressed upon the need to protect critical national infrastructure and U.K. businesses from cybersecurity vulnerabilities.

Labour MP and Minister for AI and Online Safety, Narayan claimed that U.K.’s AI Security Institute (AISI) conducted a trial of Claude Mythos and found it to be the “most capable model.”

“The U.K. has been a global leader in tracking AI cyber capabilities for 2+ years. Our testing shows accelerating capabilities and we most recently tested Anthropic’s Mythos model—the most capable model we’ve ever evaluated for cyber and the first to complete our hardest cyber-range end-to-end,” said Narayan.

Also Read: Claude Mythos: Imminent Threat or Marketing Hype by Anthropic?

Narayan’s statements come in the wake of Anthropic AI claiming that their frontier AI model poses unprecedented cybersecurity risks to critical digital infrastructure across countries owing to its advanced capabilities of finding vulnerabilities in software systems. Anthropic has refused to make public their new frontier model, deeming it “too dangerous” and made its limited “preview” version available for testing for giants like Apple, Google, NVIDIA, Broadcom among others on “freemium” basis.

The Labour MP has said that Mythos’ advanced capabilities has brought into focus the need to strengthen the Cyber Security and Resilience Bill and include “data centers to be part of Critical National Infrastructure (CNI) for the first time.”

2/ Mythos is the most capable model we’ve ever evaluated for cyber and the first to complete our hardest cyber-range end-to-end. This represents a step up in AI cyber capabilities. We’ve taken action based on our findings.

— Kanishka Narayan MP (@KanishkaNarayan) April 10, 2026

“My message to businesses: get the basics right now. Follow U.K.’s National Cyber Security Center (NCSC) guidance, get cyber essentials certification, patch vulnerabilities quickly, maintain good internet hygiene. AI capabilities are moving fast but strong fundamentals still work,’ said Narayan.

The institute is working with the National Cyber Security Centre to reduce risks. The UK government is also planning new laws to improve cybersecurity, including giving more protection to important systems like data centers. Officials are advising businesses to follow simple safety steps like updating software, fixing bugs, and maintaining strong security practices.

What is U.K AISI Cyber Range End-to-End Challenge?

UK’s AISI has developed a specialized, private 32-step enterprise network attack cyber ranges designed for high-fidelity evaluation of frontier AI agent capabilities in multi-step, end-to-end cyber attack scenarios. 

The cyber-range end-to-end challenges simulate complex enterprise networks and operational technology (OT) systems, incorporating layered configurations, identity systems, and cloud components to model real-world risks.

AISI had tested seven frontier AI models till March 2026 (before the unveiling of Claude Mythos) including Claude Opus 4.6 and claimed that none of the models were able to complete their evaluation tests.

Also Read: Anthropic Has a History of Scare Tactics: David Sacks on Claude Mythos