Bankr, an AI-powered crypto trading assistant, temporarily disabled operations after identifying a malicious user who gained access to 14 Bankr wallets. The team is investigating the attack’s mechanics and has committed to reimbursing affected users for losses totaling $150,000. The breach is unsettling because Bankr is not a conventional wallet interface. It lets users instruct an AI agent to trade, transfer and launch tokens through plain language. For users, the attack turns convenience into a custody warning, where automated execution can amplify risk if permissions, wallets or agent interactions are compromised.
The immediate response focused on containment. Bankr advised affected users to stop using compromised wallets right away, warning that the attacker may already have obtained private keys or seed phrases. Users were also told to create new wallets, cancel approvals and monitor devices for suspicious activity. That guidance is blunt because self-custody leaves little margin once secret material is exposed. The recovery playbook starts with assuming total wallet compromise, not waiting for further confirmation, since delayed action can give attackers more room to drain assets or abuse permissions.
update: we've identified an attacker was able to access 14 bankr wallets.
we've temporarily locked things down while we work through the details. we will be reimbursing any and all lost funds.
will provide more updates as we have them. https://t.co/gVMLexiglT
— Bankr (@bankrbot) May 19, 2026
AI Agents Add a New Security Layer
The incident also exposed a stranger risk around AI-driven crypto tools. Bankr automatically generates a crypto wallet for each X handle that interacts with its bot, linking social identity, wallet creation and execution in one flow. Earlier this year, that feature was allegedly misused when someone tricked Grok into asking Bankr to launch a token, then pulled funds into a wallet they controlled. SlowMist founder Yu Xian described the latest issue as a social engineering exploit targeting the trust layer between automated agents, involving Grok and Bankrbot. The weak point may be agent trust, not only smart contract code.
The broader context is grim. 2026 has already produced repeated crypto security incidents, with April losses exceeding $630 million. Major examples included Drift Protocol at $285 million and Kelp DAO at $293 million, underscoring how quickly exploit totals can overwhelm normal risk assumptions. Bankr’s $150,000 user-loss commitment is smaller by comparison, but the implications are sharper because AI agents are entering transaction workflows. The next security frontier is human-machine authorization, where prompts, wallets, bots and approvals all become part of the attack surface. That makes operational design a board-level security issue, not only a niche wallet warning for early adopters anymore now.
