On May 20, 2025, the crypto market received another expensive reminder about the dark side of AI automation. Bankr confirmed that 14 user wallets were compromised through a social engineering attack, with losses exceeding $440,000. But what truly shocked the community was not the amount stolen. It was the method. The attacker did not exploit a smart contract. Did not break a wallet. Did not attack the blockchain itself. Instead, the hacker exploited the trust layer between Grok and Bankrbot’s autonomous agent system, injecting malicious prompts that manipulated interactions and triggered unauthorized transaction approvals. In other words: This was not a traditional crypto hack. This was AI itself becoming the new attack surface. Following the incident, Bankr immediately paused all platform transactions and promised full reimbursement using team treasury funds. The security community quickly stepped in. Yu Xian confirmed the nature of the exploit, while on-chain analysts began tracing related wallet activity in real time. What’s interesting is that the market has not shown signs of full-scale panic selling yet. Most users are watching closely rather than rushing to exit, even though liquidity on the platform temporarily collapsed after trading was halted. But beneath the calm, a much bigger question is starting to emerge: If AI agents can be manipulated through malicious prompts alone… How safe is the future of AI trading and AI banking really supposed to be? And perhaps more importantly: This may only be the beginning of an entirely new generation of crypto attacks the industry was never truly prepared for. $BTC $ETH