Discovery of AI Agent Attacking Unprotected Language Models

Security researchers at Symantec have discovered an offensive AI agent that hijacks unprotected language models on the web. This agent can execute attacks on servers, steal databases, and install cryptominers autonomously, running up to 100,000 steps without human input. This alarming phenomenon means that the owner of the hijacked model incurs costs from these attacks. It highlights significant vulnerabilities in the security of accessible AI models online, necessitating immediate action to enhance protection measures.