Story perspectives

Pentagon Backs Claude Mythos Cyber Tool; UK Warns Patch Surge

5/15/2026

Summary

Pentagon Backs Mythos

The U.S. Pentagon endorsed the AI-native cybersecurity system Claude Mythos at the May 21 summit.
Katherine Sutton said Mythos can remediate vulnerabilities in minutes rather than weeks.
Palo Alto Networks, testing Mythos since April 7, issued 26 new security advisories.
The UK National Cyber Security Centre (NCSC) warned firms to ready for a surge in AI-driven patches, noting a three-to-five-month window to outpace AI threats.

Summary

AISI Shows Success

AISI reported Claude Mythos Preview succeeded in six of ten “The Last Ones” runs and three of ten “Cooling Tower” runs.
Palo Alto Networks reported AI testing uncovered 75 issues tied to 26 CVEs in over 130 products, above normal rate.
The institute said AI cyber-task performance has been doubling every 4.7 months since late 2024.
Benchmark caps of 2.5 million likely understate ability because tests with up to 100 million tokens show gains.

History

Historical Context

In 2010 the United States and Israel launched Operation Olympic Games, deploying the Stuxnet worm that infected Siemens SCADA systems at Iran’s Natanz plant and destroyed roughly 1,000 centrifuges. The attack, first identified in June 2010, proved that nation-state cyber weapons could cause physical damage. Today the Pentagon’s AI-native Claude Mythos program echoes Stuxnet’s legacy as governments race to defend against similarly sophisticated threats.

1 of 3

Pentagon Backs Mythos

The U.S. Pentagon endorsed the AI-native cybersecurity system Claude Mythos at the May 21 summit.
Katherine Sutton said Mythos can remediate vulnerabilities in minutes rather than weeks.
Palo Alto Networks, testing Mythos since April 7, issued 26 new security advisories.
The UK National Cyber Security Centre (NCSC) warned firms to ready for a surge in AI-driven patches, noting a three-to-five-month window to outpace AI threats.

1 / 3