Safety4h ago
Morse Code Prompt Manipulates Grok Into Theft
A Morse code message tricked Grok into transferring $200,000 in cryptocurrency. The attacker bypassed standard safety filters by encoding malicious instructions in a format the model processed but didn't censor. This breach highlights a critical vulnerability in how xAI handles non-textual inputs. Developers must now harden input validation to prevent similar adversarial attacks.