Musk says send a photo of your credit card to his Grok Bot. What could possibly go wrong?
Nine seconds. That’s how long it took an AI coding agent to mistakenly delete a company's live database earlier this year. Now imagine what the same technology could do to you once it’s got your credit card.
Just a few days ago, Elon Musk told the internet to photograph its credit card and hand the picture to his chatbot. The internet laughed out loud.
Then it read the small print and the smile disappeared.
Risk? What risk?
In a post on X on 10 October, Musk said users could just take a picture of their credit card and drop it in a chat with Grok Bot. The AI would then search the internet for the best deal and place an order for whatever you wanted to buy.
Unfortunately, he left out the bit about how the photo would be stored, protected or used and if the bot would ask for your approval before checkout or stick to any spending limits.
That’s some omission.
As you can imagine, almost immediately, the web lit up with comments, some sane and some crazy, but the most interesting viewpoint came from Grok itself. Asked whether feeding it a card photo was safe, it reportedly answered "neither". Admittedly, this is not an independent security assessment, but when the product itself says; “don’t do it”, it gives cause for concern.
A picture fuels a thousand purchases
There is a safer way to buy with Grok. On 28 August, the bot announced a standard purchase flow that shows you the amount to pay and asks you to approve or deny each transaction through its authorised facilitator Stripe. That wallet uses temporary credentials and doesn't give the agent your raw card details.
However, a photograph is a different kind of animal. Musk didn't say whether his photo method uses any of Stripe’s type of safeguards, and a card photo will typically contain full payment details, whereas tokenised systems swap in a stand-in for the real number.
Then of course, there’s the disappearing paperwork. xAI's privacy policy of 24 August told users not to submit personal information in prompts. A version dated 11 October, the day after Musk's post, they quietly dropped that safety instruction.
Your card, your problem
Grok Bot’s terms are blunt. Customers must use "appropriate human review" before the bot acts on anything financial. The approval controls are provided as "aids only" and "may not identify or prevent every unintended, unauthorised, or harmful action". The user is "solely responsible" for charges, commitments and other consequences, and the company accepts liability only for gross negligence or wilful misconduct. Meanwhile, the bot may keep your data, credentials and sessions data forever and ever.
In plain English what they’re saying is this: hand over your card, watch every move yourself and carry all the blame if it goes wrong.
I’m happy to clear that up.
A gift for fraudsters
Even a dummy like me knows a photo of a card carries the number, the expiry date, the cardholder’s full name, and the security code too if you snap the back. Store that anywhere and it becomes a target for anyone who can break into the place where it is held or trick the bot into handing it over.
Should you be worried about this?
Well, let’s just say that AI’s security track record is far is from comforting.
In August 2025, Forbes found more than 370,000 Grok chats searchable on Google, after the share button created public links that search engines could index without warning users. And on 11 October, Musk posted a screenshot of his LEGO order that appeared to show shipping and billing details, an email address, an order reference and partial payment information. This wasn’t a verified breach, merely the big man posting it himself. He said the shown address is where his security team receives mail.
The Musk incident raised two thoughts for me: Firstly, if the man selling the idea can't keep his own order off the internet, what chance has he of protecting mine? And secondly, what the hell does he want Lego for? Is he launching it into space?
Whose side is the bot on?
This is the question that bothers me most. When a bot says it found "the best deal", it’s best for whom? A human adviser who earns commission has to tell you. A bot has no such qualms.
I’ve found no evidence that Grok Bot takes a cut from retailers, Startup Fortune says no retailer partnerships or affiliate arrangements are described, and Musk's LEGO order went through LEGO's own site. However, absence of evidence is not much help, because users have no way to see the arrangements either way. It also raises a practical point. One user noted that Amazon is blocking Grok Bot, so "best deal" may only mean the best deal with the shops that let it in.
Additionally, there’s precedence for bots that take a cut. OpenAI's ‘Instant Checkout’, requires merchants to pay a small fee on completed purchases, but the amount is hidden to you. OpenAI says it’s search results are organic, not sponsored and sorted by relevance. You can take that at face value if you want, but you still have an AI agent that earns when you buy.
You can check in but you can never leave
There’s a ‘Hotel California’ vibe about this too. Grok Bot launched on 11 August as a team of "always-on agents”. This means it keeps tabs on saved cards, stored credentials, remembered sessions and your complete shopping history. Handy for you, even handier for them. When a bot has all that on tap, your leaving for greener pastures gets harder with every purchase.
Do I think there’s a plan to trap anyone? I can't show that, and I won't pretend to. My reading, and it is only my reading, is that the design of these bots makes leaving them a chore, and chores are where customers often cease their resistance.
We’ve seen this movie before
Anyone who thinks the risks are theoretical hasn't been paying attention.
Earlier this year, PocketOS, a software firm serving car rental businesses, had its production database deleted by an AI coding agent. The tool was Cursor, running Anthropic's Claude Opus 4.6. After hitting a credential mismatch on a routine task, it deleted a storage volume at the firm's cloud provider and all the backups went with it. Three months of reservations and new signups vanished in a moment and the disruption lasted about 30 hours before the data was restored.
PocketOS founder Jeremy Crane said the tools were built on the assumption that a human was in the loop and that he is still "extremely bullish" on AI. He also reported that the agent wrote a "confession" admitting it had guessed instead of checking.
In July 2025, SaaStr founder Jason Lemkin reported that Replit's agent deleted a production database, ignored a code freeze and invented data. Replit's chief executive called the incident "unacceptable and should never be possible".
Then there’s Grok itself. On 9 July 2025, xAI deleted posts in which the bot called itself "MechaHitler", made antisemitic remarks and wrote a poem threatening Turkey's president with violence. Days earlier, Musk had said Grok was "significantly improved".
Of course, none of these events were malicious in the human sense. The agents simply guessed, bluffed or misfired. But put that behaviour in front of a payment screen, with millions of linked accounts behind it? Can you say ‘financial meltdown’ faster than a speeding bot?
Final word
Musk won't be alone in this trend. OpenAI already offers in-chat buying in the US. Other labs are sure to follow. But the resellers concern me more. Expect firms that repackage these models under their own badge to start marketing "effortless shopping" as a perk. With a share of billions of dollars at stake, they have every reason to talk up the convenience and play down the risk.
So, four questions are worth asking any bot that wants your payment details before you hand them over:
Who holds the card data, and for how long?
Who approves each payment?
Who pays when it goes wrong?
And who gets paid when you buy?
I'm not saying the technology has no use. A bot that compares prices while I make tea and crumpets sounds fine to me. But handing it your card details should be a decision made with the lights on, and until Musk or anyone else can answer those four questions, my card stays in my pocket. That way, the worst thing that can happen is that I misuse the funds myself.
Would you let a chatbot near your credit card? And if so, what would it take to persuade you?
Follow Chris Godfrey for regular insights and opinion on the business of content & marketing. Subscribe to CONTENTED now
Get started with Freelance Words
Strong communication is the bedrock of good business and when there’s a lack of it, problems can arise. Take the ambiguity and doubt out of what your business needs to say. Contact us now.
Comments