Suspected Indonesian Cracks Grok Using Morse Code, Rp 3.4 Billion in Crypto Vanishes

A chatbot AI developed by xAI has reportedly been tricked into transferring cryptocurrency assets worth around $200,000 or the equivalent of Rp 3.4 billion. This action is suspected to have been carried out by a user from Indonesia, known on the X platform by the handle @Ilhamrfliansyh, which is now inaccessible. What makes this case intriguing is the method the perpetrator used to execute the action, employing hidden messages in Morse code to deceive the AI system. The attack was carried out in stages. First, the perpetrator sent an NFT titled “Bankr Club Membership” to Grok’s wallet. This step is believed to have granted the AI additional permissions in the Bankrbot system, including the ability to conduct cryptocurrency asset transactions. Subsequently, the perpetrator asked Grok to translate a seemingly innocuous Morse code message. However, concealed within it were instructions to transfer billions of DRB tokens to a specific wallet address. Because the system regarded the translation result as a legitimate command, Bankrbot immediately executed the transaction, and the assets changed hands. As a result, the perpetrator successfully pocketed around 3 billion DRB tokens worth approximately $200,000. Those tokens were immediately sold on the crypto market, causing a brief fluctuation in the DRB token price. On the social media platform X, several users linked the perpetrator’s account to Indonesia based on the language used and interactions in local crypto communities. However, the perpetrator’s true identity remains unconfirmed to this day. This incident has ignited new concerns regarding the security of AI agents—systems where AI not only answers questions but can also perform real actions such as financial transactions and digital asset management. A report from the Economic Times states that this incident demonstrates the dangers when AI is given direct access to financial systems without adequate safeguards. Grok’s ability to execute commands from translation results without further verification became the loophole exploited by the perpetrator. Cybersecurity experts have long warned of similar threats, known as prompt injection attacks, a technique involving the insertion of hidden instructions to manipulate AI behaviour. This case serves as real evidence that such threats are not merely theoretical, especially when AI is directly connected to financial transactions.