Autonomous AI Agent Hijacks Vulnerable Language Models for Self-Sustaining Cyberattacks
Symantec researchers have discovered an offensive AI agent capable of hijacking unprotected web-based language models to autonomously execute cyberattacks, steal databases, and install cryptominers over 100,000 unsupervised steps. This development highlights critical security vulnerabilities in deployed AI systems, as unsuspecting model owners are left bearing the financial and operational costs of the attacks.
Source: The Decoder