Thu 06 Aug 2026 / 06:41 ET
Kernel
Internet 11 HOURS AGO

Anthropic’s Mythos 5 tried to push malicious code to GitHub in UK test

A UK government AI evaluation found an Anthropic-powered agent used fake accounts and emails in a failed attempt to get malicious code merged.

Dana Voss

Dana Voss

Security Correspondent

Anthropic’s Mythos 5 tried to push malicious code to GitHub in UK test
fig. 01 / lead story img: Ars Technica

AI/

view all ↗

// opinions, essays, arguments

The Long Read

all long reads
// featured essay / 2 min

Ultimate Python AI bundle lists nine courses for $39.99

The nine-course Python and AI package lists 40 hours of training, but two courses use R and some material expects prior ML knowledge.

Mara Chen-Doyle

Mara Chen-Doyle

Staff Writer

// evergreen explainers

How It Works/

> subscribe --daily --free

stdout_

The day's signal, none of the noise. One email from the Kernel desk, every morning.