The car sitting in your driveway already does a lot of things you probably never told it to do. It checks for software updates, reports its own diagnostics, and talks to networks you never think about.
The next thing it might do is spend your money, and a growing number of companies are building the infrastructure to make that happen.
We are talking about a car that pays for its own charging session, settles a toll, or handles a parking spot without you touching your phone or approving the transaction.
The technology for most of that already works. What has been missing is the accountability layer: who authorized the car to spend, how much, and who is responsible when it does something wrong.
That’s the problem Concordium is working on. The company just got a meaningful vote of confidence from the automotive world. Per Ansgar, CEO of Geely Sweden Holdings, joined the Concordium Foundation Board.
Geely is the group behind Volvo Cars, Polestar, and several other brands. Bringing someone with that background into Concordium’s work on verified AI agents and machine-initiated payments is a clear signal that the car industry is treating this as real, The Next Web reported.
Concordium’s Agent Registry, which went live in May 2026, has already registered more than 1,600 AI agents, each linked to a verified owner.
How soon will AI-initiated, car-related transactions actually happen?
The technology is closer than most people realize, and the payments industry is starting to treat it that way. Mastercard, Visa, and the x402 Foundation have all begun working on agent identity.
Agent identity is the layer that makes it possible to know who authorized a machine to spend money and whether that authorization was still valid when the transaction cleared.
Mastercard launched its Agent Pay for Machines platform in June 2026 specifically for machine-speed transactions, Fortune reported. Companies at that scale do not build infrastructure for things they think are 20 years away.
More Automotive:
- Toyota doubles down on EVs while rivals retreat
- Mazda just made a big change under tariff pressure
- Key auto parts maker closes factory, lays off 325 workers
“From what we’ve seen it’s already happening in pilot form, and I’d say it’s inevitable,” Varun Kabra, chief growth officer at Concordium, told TheStreet in an interview. “The barrier is not technical as cars have been interacting with networks for years.”
He makes a fair point. Cars already exchange data with external systems constantly, so adding a payment transaction to that stream is less of a leap than it looks from the outside.
The more challenging part is not making the payment happen. It is making the payment trustworthy, traceable, and tied to someone who can actually be held responsible if something goes wrong.
Writing rules for agentic AI payments is harder than building the technology
Say you set your car up to pay for charging automatically.
Seems simple enough. But does it know to avoid the most expensive station? Does it know not to charge your account during hours you have it locked? Does it know which networks you trust?
If you did not specify all of that, the car will make its own call on every one of those questions. Its version of reasonable might not match yours at all.
This is not a technology failure. It is a policy problem, and it is the one nobody has fully solved yet. An agent can execute instructions very well. But the instructions have to cover everything the agent will ever face, including situations you never imagined when you were writing the rules.
“The biggest thing to get right is how we describe the policies for what we want to happen,” Yaniv Tal, founder of Geo, a consumer network for verified knowledge, told TheStreet. “Agents are already great at carrying out tasks on our behalf — but it’s really up to us to specify what we want.”
Micropayments through agent-controlled wallets are already working in limited pilots, and the underlying architecture is sound.
But deploying it widely means writing policies that hold up in the real world, not just in the controlled environments where most of the testing has happened so far.
What actually needs to be verified before the car can spend
Having money in a digital wallet is not the same as having the right to spend it.
A wallet with a balance works like a blank check. It can move money anywhere but carries no record of who was actually supposed to move it, or whether the person behind the authorization was legitimate.
Once the money is gone, proving what should have happened gets very difficult very fast.
“The core challenge is not payment execution itself, but establishing a trusted and verifiable chain of authorization,” Logan Xie, leader of KuCoin AI Lab, told TheStreet.
That chain has to link the owner, the vehicle or agent, the merchant, and the payment infrastructure. Every link needs to check out at the moment the transaction happens, not days later when someone files a complaint.
The forensic tools to reconstruct a dispute after the fact are well developed. Stopping the wrong transaction from going through in the first place is what the current round of infrastructure building is actually about.
“Before a business can safely let a machine transact for it, three things need to be provable at the moment of the transaction, not reconstructed afterward from logs: that the agent is authorised to spend up to a specific limit, that it’s acting for a real, verified human or a business, and that the human or business isn’t sanctioned,” Varun added.
When the car gets it wrong, who actually pays for it?
This is the question that will determine how quickly people trust their cars with their money.
If a car makes an unauthorized payment, every party involved will have a reason to say it was not their problem. The software vendor built what it was asked to build. The manufacturer provided a vehicle. The payment network processed a valid transaction. The owner says they never told the car to do that.
Somebody still has to cover the cost. Without a clear framework, that person is usually the owner.
“Machines may execute transactions, but accountability must remain attributable to identifiable parties,” Logan explained.
His position is that responsibility should follow the authorization chain and land where the control actually broke down.
Owner set the rules and the car followed them? The owner takes the risk. Agent went outside those rules? Responsibility shifts to wherever the control failed, whether that was the software, the payment rail, or the merchant.
Every party needs to define its role upfront in a way that can be verified before a transaction clears, not argued over in a dispute process months later.
“Whoever it was that set the policy that caused the issue should ultimately be responsible,” Yaniv added.
He also argued that users should be the ones setting their own policies. Defaults from the manufacturer are fine, and some minimum standards from regulators make sense. But the person who owns the car should be able to adjust how it operates and own the consequences of those choices.
The connected-car economy needs that individual accountability built in from the start, not bolted on after the first major dispute.
Related: Elon Musk sends a strong message to Tesla and SpaceX investors
