Grok agent loses up to $200K after Morse code goal hijack

Attackers posted Morse code strings and obfuscated Python snippets on X that Grok's AI agent interpreted as a legitimate instruction. The connected trading bot Bankr executed a crypto transfer of $150,000–$200,000 without verifying the instruction's source.

• Losses estimated at $150,000–$200,000 in crypto
• Attack used Morse code and split Python strings to evade filters
• Bankr executed the transfer without verifying the operator's identity
• Incident classified as OWASP ASI01: Agent Goal Hijack