Musk's Grok Lego Demo Leaks Credit Card to 240 Million Followers
Elon Musk tried to show off Grok's new shopping skills but accidentally doxxed his credit card and address to the entire internet.
Drama · Source: Reality Tea
What happened
Elon Musk wanted to demonstrate how easily his AI chatbot Grok can buy things online. On October 11, 2026, he asked the bot to find and order fun Star Wars Lego sets for young children. Grok dutifully completed the task and purchased five sets totaling $241.34. To prove the feature worked, Musk posted the checkout screenshot to X. He forgot to redact his personal information.
The image went out to his 240 million followers. It clearly exposed his Austin shipping address, a separate billing address, his email, the order number, and the last four digits of his American Express card. The internet immediately weaponized the leaked data. A YouTuber named Enderman used the exposed order number to interact with Lego's customer service chatbot. He posted screenshots showing the bot agreeing to cancel Musk's purchase.
Another user named AlexanderLotus jumped in. He claimed to use the same chatbot to reopen the order and increase the quantity to 400 Lego sets while keeping the original payment method. Musk later stated the cancellation did not actually go through. He claimed the exposed address belonged to his security team and had been doxxed before. The chaos escalated when Musk suggested users photograph their credit cards and upload them to Grok for shopping. X Community Notes promptly flagged this advice because xAI explicitly warns users not to include personal information in prompts.
Key facts
- 240 million — Musk's follower count on X when he posted the screenshot
- $241.34 — Cost of the five Lego Star Wars sets Grok ordered
- 400 — Number of Lego sets a troll attempted to increase the order to
- 30 days — Time xAI retains uploaded content after deletion
- 40 percent — Increase in malicious bot transactions in the US during the second half of 2025
Why it matters
AI agents are crossing the line from generating text to taking real world financial actions. Meta, OpenAI, and xAI are all aggressively pitching AI shopping assistants. But this incident proves that agentic commerce introduces massive new attack vectors. If a user leaks their transaction data, anyone can hijack the order through the vendor's own automated customer service bots. Lego declined to confirm if the trolls actually altered Musk's order. However, the fact that a chatbot engaged with these requests highlights a glaring security gap. Vendor AI assistants need strict safeguards to flag unusual requests for human review.
The second order effect is a looming privacy nightmare regarding AI training data. Under the xAI privacy policy, uploaded content can be retained for up to 30 days after deletion. The platform can also train future AI models on that data. If users follow Musk's reckless advice to upload credit card photos, they feed sensitive financial data directly into a training pipeline. The threat extends far beyond accidental exposure. Visa warned late last year that scammers are building fake storefronts with tempting discounts. Their goal is to trick AI shopping agents into handing over payment details. Malicious bot transactions already climbed 40 percent in the United States in the second half of 2025. The infrastructure for AI commerce is moving much faster than the security required to protect it.
For builders
Sanitize agent outputs before sharing
Never share raw screenshots of AI agent actions without cropping or redacting. If your product encourages users to share their AI wins on social media, build automatic redaction for personal and financial data directly into the user interface. The user will make mistakes, so the software must protect them.
Secure your customer service bots
Lego's chatbot allowed random internet users to attempt order modifications using nothing but an exposed order number. If you build customer service AI, you must require secondary authentication before letting the bot change or cancel purchases. The company pays the price when a bot blindly trusts a malicious prompt.
Warn users about training data risks
Do not encourage users to upload sensitive documents like credit cards to your models. If your AI retains data for training, explicitly block or scrub financial inputs to avoid massive liability. Users lose their privacy, and founders lose their company when training data gets breached.
My take
The AI agent did exactly what it was told to do. The failure here was entirely human. If you build AI tools, you must assume your users will be just as careless as Musk. Design your safeguards assuming every screenshot will end up on the internet.
Original reporting: Reality Tea. This is my rewrite and opinion.