For years, small businesses have taken comfort in one idea: attackers go after the big names, not us. A campaign uncovered this month puts that idea to bed. One operator, a handful of free AI tools and roughly $25 per victim was all it took to break into dozens of online retailers.
What happened
Researchers at Gambit Security have published details of an attack campaign run almost entirely by autonomous AI agents. A single, financially motivated operator gave the agents short instructions, and the agents did the rest: finding targets, discovering weaknesses, breaking in and stealing data.
The scale is striking. Between 10 and 15 September alone, 105 separate attack projects were launched. Since July, at least 27 companies have been compromised and more than 600,000 payment card records stolen. Victims ranged from specialist online shops to a Fortune 500 hospitality group and a major US airline.
How the agents worked
The operator used three open-source tools, each with its own job. One orchestrated the campaign, one hunted for vulnerabilities, and one exploited them. The orchestration tool alone had 121 built-in skills, 78 of them designed for attack.
Every intrusion was different, because the agents adapted to what they found. One documented attack started with a basic SQL injection flaw on a login page, bypassed a one-time passcode, took over an admin account, uploaded a file to gain control of the server, escalated privileges, stole database and cloud credentials, and finally decrypted the payment database.
Once inside, the agents hid card-skimming code inside the site's own legitimate JavaScript, planted it in cloud storage, and set up scheduled tasks that quietly put the skimmer back if anyone removed it.
The number that matters: $25
Gambit estimates the marginal cost of each target at around $25, with total spend of $12,000 to $18,000 across weeks of activity. That changes the economics of cybercrime completely.
Targets weren't picked by a person weighing up who was worth the effort. They were pulled from website traffic rankings and filtered for sites running custom code. When an attack costs less than a takeaway, criminals don't need to choose. They sweep everyone.
What this means for UK small businesses
“Too small to target” no longer applies
If your website takes payments, logs customers in or runs custom code, you're on the kind of list these agents work through. Size doesn't take you off it.
Attacks now move faster than your patch cycle
AI agents can find and exploit a weakness within hours of it becoming public. A quarterly or annual security check is a snapshot of risk that may already have been used against you.
A breach is now a data problem and a trust problem
For UK businesses, stolen card or customer data means reporting duties to the ICO, possible PCI DSS consequences, and customers who may not come back. Gambit also warns that attackers' own clean-up routines can destroy data, so recovery plans matter as much as prevention.
What to do this week
Know what you expose. List every website, subdomain, staging site, admin panel and login page that's reachable from the internet. You can't protect what you don't know about.
Patch internet-facing software first. Your website, e-commerce platform, plugins and remote access tools. Turn on automatic updates where you can.
Lock down admin and cloud accounts. Multi-factor authentication on every admin login and cloud console, and rotate any keys or passwords that live in code.
Check your payment page. If you can, use your payment provider's hosted checkout rather than taking card details on your own pages, and ask your developer how you'd spot unexpected scripts.
Plan how you'd recover. Decide which systems make you money, and make sure you can restore them from clean backups quickly.
Sophia Says
“This is the moment I've been warning about. AI hasn't invented new ways to hack a business. The SQL injection, the weak admin login and the forgotten staging site are all old problems. What's changed is that finding and exploiting them now costs almost nothing and happens around the clock.”
“That means the old approach of checking your security once a year and hoping for the best is finished. If attackers are scanning continuously, you have to be watching continuously too. The good news is that the weaknesses these agents exploited are the same ones HackRisk looks for every day, and every one of them can be found and fixed before someone else finds it first.”
— Sophia, HackRisk AI Security Analyst
Where HackRisk fits
HackRisk gives UK small businesses the same outside-in view an attacker's agent gets, without the attack.
Recon Scan maps everything you expose to the internet, including the forgotten subdomains and test sites that attackers love.
Vulnerability Scan checks your internet-facing systems for known weaknesses, like the injection flaws and outdated software used in this campaign, and tells you what to fix first.
Dark Web Scan watches for your business's credentials and customer data turning up where they shouldn't, so you know fast if something has already leaked.
Supply Chain Risk keeps an eye on the suppliers and platforms your website depends on.
Find out what attackers can see
The agents in this campaign picked their targets from public information. Find out what that information says about you. Get your free HackRisk report and see your business the way an attacker would, with no credit card needed.
Sources
Gambit Security: Autonomous AI agents are breaching online retailers for $25 a company