Story perspectives

CISA Deploys Anthropic’s Mythos AI, Finds Federal Code Flaws

7/8/2026

Summary

CISA Audits

CISA is using Anthropic’s unreleased AI model Mythos to audit federal code repositories for cybersecurity flaws.
The Attack Surface Evaluation team is conducting audits across multiple systems, with Reuters reporting many software vulnerabilities uncovered.
The NSA began using Mythos in April despite supply-chain risk designations, while a judge blocked the Pentagon’s risk designation in March and agencies such as Treasury, State and HHS are phasing out Anthropic tools.

Summary

Anthropic Export Talks

Howard Lutnick, Sean Cairncross, Anthropic co-founder Tom Brown and policy head Sarah Heck met to discuss lifting the export ban on Anthropic’s Fable 5 and Mythos 5 models.
Anthropic shut down all access to the Fable 5 and Mythos 5 models to meet the new regulations.
Breitbart social media director Wynton Hall called AI a “landmine” and, in *Code Red*, said the MAGA movement should shape AI policy.

History

Historical Context

In 1996 President Bill Clinton signed Executive Order 13026, moving commercial encryption from the Munitions List to the Commerce Control List and enabling the newly-signed Wassenaar Arrangement—joined by 39 countries—to coordinate dual-use export rules for cryptography. Today, U.S. agencies continue to apply the framework as they consider future AI model restrictions.

1 of 3

CISA Audits

CISA is using Anthropic’s unreleased AI model Mythos to audit federal code repositories for cybersecurity flaws.
The Attack Surface Evaluation team is conducting audits across multiple systems, with Reuters reporting many software vulnerabilities uncovered.
The NSA began using Mythos in April despite supply-chain risk designations, while a judge blocked the Pentagon’s risk designation in March and agencies such as Treasury, State and HHS are phasing out Anthropic tools.

1 / 3