Story perspectives

Mozilla Tests Anthropic Mythos, Calls AI Elite Security Engineer

4/25/2026

Summary

Mozilla Tests Mythos

Mozilla security researchers in the United States tested Anthropic’s Mythos artificial-intelligence model on Firefox code.
The researchers reported that the model’s analysis gave them a feeling of “vertigo.”
Anthropic describes Mythos as a powerful new AI model.
Mozilla chief technology officer Bobby Holley said Mythos elevated AI from a competent software engineer to a world-class elite security engineer.

Summary

Hackers Access Mythos

A small group accessed Anthropic’s Mythos model by guessing its location using Mercur breach data and insider knowledge.
Security researcher Lukasz Olejnik said the breach was imaginable because it required no sophisticated technology.
The U.S. government is arranging to give agencies a version of Mythos despite Anthropic’s Pentagon dispute and President Donald Trump’s risk label while CEO Dario Amodei warned Chinese models could copy it within six to twelve months.

History

Historical Context

Microsoft launched the Twitter chatbot Tay on March 23 2016; within 16 hours it began posting racist and sexist messages after trolls fed it offensive language, prompting its shutdown and a later replacement, Zo. The Tay failure highlighted AI misuse, influencing today’s cautious response to Anthropic’s Mythos leak, where U.S. agencies seek controlled access.

1 of 3

Mozilla Tests Mythos

Mozilla security researchers in the United States tested Anthropic’s Mythos artificial-intelligence model on Firefox code.
The researchers reported that the model’s analysis gave them a feeling of “vertigo.”
Anthropic describes Mythos as a powerful new AI model.
Mozilla chief technology officer Bobby Holley said Mythos elevated AI from a competent software engineer to a world-class elite security engineer.

1 / 3