Letting an AI spend money for you sounds convenient right up until you realize that “convenient” is usually how bad ideas get smuggled into your life.
That’s what jumped out at me about the news that OpenAI has introduced a “ChatGPT Wallet,” a payment method meant to let ChatGPT make purchases on a consumer’s behalf. From what’s been shared publicly, it’s designed to be secure and to plug directly into card networks so the AI can help complete transactions, not just talk about them. This is part of a bigger push to make ChatGPT useful for real errands, not just writing and questions.
On paper, it’s the obvious next step. People already ask AI what to buy, where to buy it, and how to compare options. The gap between “recommend” and “purchase” is just a few clicks. The temptation is to remove those clicks and call it progress.
But those clicks are doing more work than we admit. They’re friction. They’re the moment you pause and think, “Do I actually want this?” They’re where your judgment gets one last vote.
If ChatGPT can pay, then it stops being just a tool that advises you. It becomes a tool that acts for you. And once something can act, the standard for trust changes. A wrong answer is annoying. A wrong purchase is a mess.
Imagine a normal night: you’re tired, you tell ChatGPT to reorder a household item, and it picks a listing that looks right but isn’t. Maybe the size is wrong. Maybe it’s a third-party seller. Maybe it’s a “similar” product that’s cheaper for a reason. Now you’re dealing with returns, customer support, and the weird feeling that you didn’t exactly choose this, but you’re still on the hook.
Or imagine something more sensitive. Say you ask it to book travel. It might pick a flight with a brutal layover because it optimized for price. Or it might miss a detail you care about—like arrival time for a wedding. When humans book travel, we make tiny tradeoffs based on context we barely explain, even to ourselves. AI can only act on what it can “see” in your request and what it’s allowed to assume. That gap is where disappointment lives.
The part that makes me most uneasy isn’t that it will make mistakes. Everything makes mistakes. It’s that spending money is a high-trust action, and the incentives around spending money are always ugly.
Even if OpenAI builds this with real security, “secure” doesn’t automatically mean “aligned with you.” Secure just means the payment can go through safely. It doesn’t mean the purchase was wise, or fair, or in your interest. A wallet that works perfectly can still steer you into dumb choices faster than you can notice.
And yes, I understand the upside. For people who are busy, overwhelmed, or juggling a million things, an AI that can handle routine purchases could be a relief. If you have limited mobility, if you’re caring for family, if you hate dealing with checkout forms and password resets, this could genuinely help. It could also reduce small daily stress. That matters.
But I also think it will quietly change people’s habits in a way that’s hard to reverse. If the AI can buy, people will start asking less and delegating more. The muscle of comparison shopping weakens. The habit of second-guessing fades. You’ll default to “just handle it” because that’s what tools are for. Then one day you’ll look at a charge and think, “Why did I buy this?” and the answer will be, “Well, you told it to.”
There’s another uncomfortable angle: when a system becomes the shopping doer, it becomes the shopping gatekeeper. It decides what options you see and what options you don’t. It decides what counts as “best.” If it’s connected into payment rails, it’s not just a chatbot anymore. It’s part of commerce.
That raises the question of whose preferences are being optimized. Yours, obviously, is the promise. But commerce is full of nudges: cheaper for the seller, better margins, partnerships, defaults that happen to be profitable. I’m not claiming any of that is happening here. I’m saying the pressure will be there. The moment an AI can complete a purchase, people will try to influence what it buys.
And we haven’t even touched the human side of accountability. If the AI buys the wrong thing, who fixes it? You. If it subscribes you to something you didn’t mean to subscribe to, who cancels it? You. If it makes a purchase at the wrong time and your budget gets tight, who eats that stress? You.
So this feature is either going to come with strong guardrails—or it’s going to become one more system that makes spending frictionless and regret expensive. Maybe you’ll be able to set strict limits, approvals, and categories. Maybe every purchase requires a final “yes.” But if that’s the case, we’re back to the clicks we removed in the first place.
I can already hear the counterargument: people hand their card to apps all the time, and it’s fine. Auto-pay is everywhere. One-click buying exists. True. The difference is that those systems don’t “think.” They don’t interpret a vague request like “get me a good deal on running shoes” and turn it into a specific charge.
Once you let interpretation trigger payment, you’ve created a new kind of risk. Not hacking risk. Meaning risk. The risk that your intent gets translated into action badly, quickly, and repeatedly.
So I’m not against AI helping with purchases. I’m against pretending that “making it secure” solves the real problem, which is judgment. The real question is how much of your judgment you want to outsource, and what you’re willing to lose in exchange for saving a minute.
Where should the line be between an AI that recommends and an AI that spends your money?