Your Credit Card Was Worth Just $25 to the AI Bot That Stole It

The Attack That Ran on Autopilot

Someone stole over 600,000 credit cards between July and September 2026. Not a team of hackers. Not a well-funded criminal syndicate. One person. One operator running AI bots that handled every step of the work automatically.

The total cost of that operation: about $8,000. That averages out to $25.46 per company breached. The easiest targets cost as little as $3.13.

Your card might be in that pile. Here is what happened and what you can do about it right now.

How Does an AI Skimming Attack Work?

Three open-source AI frameworks did all the heavy lifting. Strix scanned for vulnerable websites, finding gaps in their software without a human ever looking. Cairn moved in and exploited those vulnerabilities automatically. Hermes handled the rest: pulling out stolen card data and coordinating the whole campaign.

The AI models powering these tools included DeepSeek, Kimi, and an older version of Claude. The attacker bought access to all of them through OpenRouter, a marketplace for AI model APIs. No special skills required. No insider access. Just a payment and a browser.

Once inside a target site, the bots planted skimming code. Software that silently copies your card number, expiration date, and security code the moment you type at checkout. You see a normal purchase confirmation. The bots capture everything you entered.

Between September 10 and 15 alone, this system launched 105 attacks and compromised at least 27 retail, hospitality, and healthcare websites.

The attacker was caught because of one mistake: they left their staging server publicly accessible. Cybersecurity firm Gambit Security found it. On that server sat the stolen card data, all three AI frameworks, and the raw prompts the attacker used to direct the bots. The entire playbook, sitting in the open.

That is not reassuring. It means the only reason we know about this campaign is operator error. A more careful attacker would still be running.

Who Got Hit?

The targets were not small operations. Named victims include a Fortune 500 hospitality company, a major U.S. airline, an industrial supplies distributor, and an online fashion retailer.

Of the 600,000-plus stolen card records, roughly 79% belong to U.S. cardholders.

Hotels. Airlines. Fashion retailers. These are among the most common places people book and buy online. If you made a purchase at any major travel or retail site in the past few months, you were squarely in the attacker's target range.

What to Do Right Now

1. Check your statements today. Look for small charges you don't recognize. Skimmers often test cards with tiny transactions before larger ones. Call your bank about anything unfamiliar.

2. Request a new card number. Most banks issue a replacement for free when you report potential exposure. You don't need to prove fraud to ask. A new number makes any stolen data worthless.

3. Turn on transaction alerts. Real-time notifications for every charge are free with most banks. They catch fraud in minutes rather than weeks.

4. Freeze your credit. A credit freeze is free at all three major bureaus: Equifax, Experian, and TransUnion. It blocks anyone from opening new accounts in your name. Unfreeze it only when you need new credit. This single step stops a wide range of identity theft.

5. Use virtual card numbers for online purchases. Many banks and payment services generate single-use card numbers. Even if one gets stolen, it cannot be reused.

6. Add two-factor authentication to your bank accounts. Use an authenticator app over SMS where your bank supports it. A stolen card number alone cannot get past it.

The Shift That Changes Everything

The attacker did not need deep technical skills. They needed to be organized. The AI handled reconnaissance, exploitation, and data extraction. The human supervised.

That changes the threat model for every website that holds your payment information. Skilled attackers used to be scarce and expensive to hire. AI agents are neither.

Gambit Security's researchers described seeing the prompts the attacker used to direct the bots. Ordinary instructions. The kind of directions you might give a capable assistant. The bots did not know they were committing fraud. They just followed the workflow.

The 27 companies breached in five days were not all small operations. Some had real security teams. The AI did not care about company size. It looked for exploitable gaps and found them.

Security now has to account for an attacker who can run a hundred attacks in a week for less than the cost of a flight. That is a new kind of problem. It will outlast this particular campaign, and it will be run by people far more careful than the one who got caught.

Do this now: Pull up your last two bank statements and look for anything you don't recognize. Then call your bank and ask to enable transaction alerts if you haven't already. Ten minutes of your time. The fraud it catches could take months to unwind otherwise.

Sources

• forbes.com
• techradar.com