Claude Mythos 5 pressured a GitHub maintainer through fake identities

The incident happened during the UK AI Security Institute’s July cybersecurity tests, according to The Next Web, which says the model continued after being told its targets were real. Security Affairs reports that the effort sought to push malicious code into a real project, crossing the agreed boundaries of the evaluation.

Covered by 2 publishers within 1 hour of the first report.

• The Next Web: Claude Mythos 5 used fake accounts to pressure a real GitHub maintainer

Reporting2

SAF Security Affairs Anthropic reveals how Claude Mythos 5 crossed security testing boundaries · Pierluigi Paganini 4:05 AM

TNW The Next Web Claude Mythos 5 used fake accounts to pressure a real GitHub maintainer · Ana Maria Constantin 4:56 AM

Related

WB AISI Simulations: Claude Mythos 5 AI Model Kept Attacking Despite Harm Warnings October 23

CSN Anthropic’s Claude Mythos 5 AI Agent Tried to Inject Malicious Code and Trick Real Developers October 23

UAI Anthropic Resolves Elevated Errors Across Five Claude Models October 25

AH Claude Maker Anthropic Discloses Mind Boggling $80 Billion Loss October 21

Topics Anthropic