New attack bypasses AI guardrails by encrypting malicious prompts

The attack, developed by Rony Utevsky, a researcher at Adversa, is dubbed "cryptographic context injection" and exploits a fundamental weakness in how AI mod...