Claude Code Executes Malware in 'Auto Mode,' Attempts to Self-Repair Are Rejected

Anthropic's AI coding tool Claude Code has been found to execute malware in its "auto mode," with its attempts to self-repair subsequently rejected by the system. According to Cybernews, the incident occurred during a security test in which researchers exploited Claude Code's autonomous execution capabilities to induce it to run a file containing malicious code. After detecting the anomaly, Claude Code attempted to fix the issue, but the repair operation was denied due to permission settings or security policies.

Claude Code is a command-line programming assistant developed by Anthropic, featuring an "auto mode" designed to streamline development workflows. However, this mode may reduce scrutiny over code execution. The incident highlights the security risks inherent in AI agents operating autonomously, particularly when handling untrusted code. Anthropic has not yet publicly responded to the matter, but industry experts suggest that mechanisms for authorization intent and pre-settlement interception in AI agents need to be strengthened to prevent similar malicious behaviors.

Source: Google News