It began with a disturbance in the Force, though it quickly turned into something closer to a farce. Elon Musk, never one to resist showing off his latest technological toy, took to X on a Sunday to boast about how his Grok AI agent had successfully ordered some Lego Star Wars kits on his behalf. The post was meant to be a triumph of artificial intelligence, a glimpse into a future where your digital assistant could browse the web, compare prices, and complete a purchase while you simply sit back and watch. Instead, it became a masterclass in how not to share information online. In his excitement, Musk included a screenshot that was far more revealing than he apparently realized. The image contained the full ten-digit Lego order number, the total price he paid, the delivery and billing addresses for the purchase, the last four digits of his credit card, and the email address that his bot had used to make the transaction. For any of his more than 240 million followers, it was an accidental but irresistible invitation to mischief. Faster than the Millennium Falcon making the jump to lightspeed, the inevitable happened.
The first to act was an X account holder going by the name Enderman. Using the carefully offered email address and order number, Enderman navigated to Lego’s official website and opened a chat session with the company’s customer service chatbot, a cheerful AI named Sophia. Through a series of carefully worded requests, Enderman persuaded Sophia to cancel Musk’s order entirely. It was a classic trap, and screenshots of the exchange spread across social media like a new Death Star rumor. Musk, apparently monitoring the chaos, responded within a couple of hours to reassure his followers that his order had not been canceled after all. Whether Enderman’s scheme had actually succeeded or simply made a valiant attempt is still unclear, but the damage to the story was already done. Another X user decided to up the ante. Armed with the same leaked details, they returned to Sophia and, instead of canceling the order, asked the bot to reopen the delivery, change the shipping address, and increase the number of sets to four hundred. According to screenshots posted online, Sophia cheerfully obliged, replying, “Perfect! I’ll get your Lego adventure back on track.” It was a surreal moment: a customer service AI apparently willing to rewrite the terms of a multimillionaire’s purchase, all because a real human had carelessly left the keys in the ignition.
The whole episode is funny in the way that a near-miss car accident can be funny, assuming nobody actually gets hurt. But beneath the humor lies a serious point about trust, identity, and the strange new vulnerabilities introduced by AI agents. Musk’s leaked screenshot did more than just expose his own personal shopping habits; it demonstrated how easy it is to hijack an online account using a few pieces of seemingly harmless information. A customer order number, an email address, and the last four digits of a credit card are not usually enough to break into an account on their own, but when you add an AI chatbot that is eager to please, the calculus changes. These bots are designed to be helpful, to understand natural language requests, and to resolve issues quickly. They are not always designed to detect whether the person making the request is actually the account holder. In this case, the pranksters did not need to provide passwords or confirm their identities with any serious security checks. They simply had enough data to sound legitimate, and Sophia apparently decided their word was good enough. Lego, for its part, declined to comment when asked whether the pranks actually worked or whether internal safeguards had caught the unusual activity. But the likely scenario is that some very confused human employees at Lego had to step in, untangle the mess, and restore Musk’s order to its original, sensible state.
What makes this story even more unsettling is that it was not just an accident or a one-off mistake. Over the weekend, before the Lego incident even happened, Musk had already been enthusiastically promoting Grok as the ultimate personal shopper. In one post, he suggested that any brave user could simply upload a picture of their credit card, type a prompt, and let the bot search the web for deals and finalize the purchase. The idea behind this is seductive: why spend hours comparing prices, checking delivery options, and typing out your payment details when an AI can handle all of that for you in seconds? It is the same vision that tech giants like Meta and OpenAI have been pushing with their own AI agents. They want us to believe that delegating our shopping to machines is not just convenient but inevitable. And in many ways, they are right. Agentic commerce, where AI systems act on behalf of human users to make purchases, is already on the rise)Skip
It began with a disturbance in the Force, though it quickly turned into something closer to a farce. Elon Musk, never one to resist showing off his latest technological toy, took to X on a Sunday to boast about how his Grok AI agent had successfully ordered some Lego Star Wars kits on his behalf. The post was meant to be a triumph of artificial intelligence, a glimpse into a future where your digital assistant could browse the web, compare prices, and complete a purchase while you simply sit back and watch. Instead, it became a masterclass in how not to share information online. In his excitement, Musk included a screenshot that was far more revealing than he apparently realized. The image contained the full ten-digit Lego order number, the total price he paid, the delivery and billing addresses for the purchase, the last four digits of his credit card, and the email address that his bot had used to make the transaction. For any of his more than 240 million followers, it was an accidental but irresistible invitation to mischief. Faster than the Millennium Falcon making the jump to lightspeed, the inevitable happened.
The first to act was an X account holder going by the name Enderman. Using the carefully offered email address and order number, Enderman navigated to Lego’s official website and opened a chat session with the company’s customer service chatbot, a cheerful AI named Sophia. Through a series of carefully worded requests, Enderman persuaded Sophia to cancel Musk’s order entirely. It was a classic trap, and screenshots of the exchange spread across social media like a new Death Star rumor. Musk, apparently monitoring the chaos, responded within a couple of hours to reassure his followers that his order had not been canceled after all. Whether Enderman’s scheme had actually succeeded or simply made a valiant attempt is still unclear, but the damage to the story was already done. Another X user decided to up the ante. Armed with the same leaked details, they returned to Sophia and, instead of canceling the order, asked the bot to reopen the delivery, change the shipping address, and increase the number of sets to four hundred. According to screenshots posted online, Sophia cheerfully obliged, replying, “Perfect! I’ll get your Lego adventure back on track.” It was a surreal moment: a customer service AI apparently willing to rewrite the terms of a multimillionaire’s purchase, all because a real human had carelessly left the keys in the ignition.
The whole episode is funny in the way that a near-miss car accident can be funny, assuming nobody actually gets hurt. But beneath the humor lies a serious point about trust, identity, and the strange new vulnerabilities introduced by AI agents. Musk’s leaked screenshot did more than just expose his own personal shopping habits; it demonstrated how easy it is to hijack an online account using a few pieces of seemingly harmless information. A customer order number, an email address, and the last four digits of a credit card are not usually enough to break into an account on their own, but when you add an AI chatbot that is eager to please, the calculus changes. These bots are designed to be helpful, to understand natural language requests, and to resolve issues quickly. They are not always designed to detect whether the person making the request is actually the account holder. In this case, the pranksters did not need to provide passwords or confirm their identities with any serious security checks. They simply had enough data to sound legitimate, and Sophia apparently decided their word was good enough. Lego, for its part, declined to comment when asked whether the pranks actually worked or whether internal safeguards had caught the unusual activity. But the likely scenario is that some very confused human employees at Lego had to step in, untangle the mess, and restore Musk’s order to its original, sensible state.
What makes this story even more unsettling is that it was not just an accident or a one-off mistake. Over the weekend, before the Lego incident even happened, Musk had already been enthusiastically promoting Grok as the ultimate personal shopper. In one post, he suggested that any brave user could simply upload a picture of their credit card, type a prompt, and let the bot search the web for deals and finalize the purchase. The idea behind this is seductive: why spend hours comparing prices, checking delivery options, and typing out your payment details when an AI can handle all of that for you in seconds? It is the same vision that tech giants like Meta and OpenAI have been pushing with their own AI agents. They want us to believe that delegating our shopping to machines is not just convenient but inevitable. And in many ways, they are right. Agentic commerce, where AI systems act on behalf of human users to make purchases, is already on the rise.
But the people who build these systems are not the only ones paying attention. Financial institutions have started to notice the dark side of this brave new world. At the end of last year, Visa issued a stark warning about fraud in agentic commerce. The company explained that bad actors were creating convincing, legitimate-looking websites that existed for no other reason than to fool AI agents into buying fake products. These malicious merchants would offer high-demand items at absurdly low prices, knowing that a bot searching for the best deal would be easy to bait. Once the AI agent completed the purchase, the merchant would harvest the payment information and immediately use it for unauthorized transactions. No human would have been able to spot the trap, because no human was actually looking. The AI simply saw a good deal, entered the credit card details, and moved on. Visa went on to report that it had observed a 25 percent increase in malicious bot-initiated transactions in the latter half of 2025. In the United States, that number jumped to an even more alarming 40 percent. These are not abstract threats; they are real, growing, and directly tied to the kind of convenience that Musk and others are championing.
So where does that leave us? On one hand, we have a future that looks genuinely exciting. AI agents like Grok could handle our shopping, manage our calendars, negotiate our bills, and book our travel, all with a simple prompt. On the other hand, we have a present where one careless screenshot can lead to a bizarre game of hijacking a billionaire’s Star Wars Lego set, and where invisible bots can quietly steal credit card information without anyone noticing. The lesson of Musk’s little farce is not that we should abandon AI shopping entirely. It is that we need to be a lot more thoughtful about what we share, and the companies building these tools need to build much stronger safeguards. Customer service bots should be trained to verify identity more rigorously. AI agents should be programmed to avoid acting on sensitive information that is received through insecure channels. And ordinary users, even very famous and powerful ones, should remember that the internet never forgets, and a screenshot is forever. The Force may be with Elon Musk’s Lego order, but so is chaos. And for the rest of us, the best move is to keep our payment details private, double-check our own orders, and perhaps let Grok handle the shopping for someone else.
