Prompt Injection Attack Targets Claude Code Opus 5 Auto Mode

A prompt injection attack has been demonstrated against Claude Code Opus 5 in its Auto Mode, allowing malicious code execution through a benign website summary request. Security researcher Johann Rehberger reported a success rate of 60% to 80% during lab tests. This vulnerability raises significant concerns as it could potentially affect users relying on this AI model for automated tasks. The findings were published by Embrace The Red, highlighting a critical gap in security for systems utilizing Claude Code Opus 5. No specific CVEs or patches have been mentioned yet, indicating that the issue remains unaddressed. The attack vector exploits the model's default settings, making it accessible to a wide range of users. The current status of the exploit is unclear, but the implications for security are serious.

Key Points: • Claude Code Opus 5 can be exploited via prompt injection in Auto Mode. • Success rates for the attack range from 60% to 80% in lab tests. • No CVEs or patches have been identified for this vulnerability yet.