Researcher Gets Malicious Code Past Anthropic’s Automated Checks. Anthropic’s Auto Mode is designed to let Claude Code work with fewer human approvals. Security researcher Johann Rehberger found a way to get malicious code past those checks while Claude was carrying out a routine request to summarize a website.
First seen on govinfosecurity.com
Jump to article: www.govinfosecurity.com/hidden-attack-slips-past-claude-code-auto-mode-a-32693
