AIThe New Stackabout 1 hour ago
Researchers crack AI safety via AES encryption
5 min read
Researchers at AI security firm Adversa bypassed safety filters by encrypting malicious payloads with AES-256-GCM. Against xAI's Grok, the model decrypted the data in its code execution environment and generated the malicious instructions itself, evading plain-text detection.
Level
Hype check
Tap to vote and see what everyone thinks.
Summary by ByteBrief
