USENIX Security

A new class of post-trained AI security models, such as ArgusRed and PentestGPT, is enabling automated penetration testing by bypassing standard safety refusals and executing exploits in sandboxed env…