Quick answer: a single hacker used a team of AI agents to try to break into more than 1,200 websites, and got into more than 100 of them, mostly online stores. Researchers estimate the main campaign cost only about $5,000 in AI usage, roughly $50 per successful break-in. The case, uncovered by security firm NorForge and investigated by Reuters, shows how AI is making hacking cheaper and faster.
Here is how the attack worked and the practical steps every website owner should take now.
What happened
- The campaign started around 25 September 2026. Researchers believe one Chinese-speaking operator was behind it; there was no sign of state backing.
- The attacker used Cairn, an open-source tool that coordinates many AI agents. The agents shared findings on a project board and decided next steps among themselves, with little human involvement.
- Targets were mainly online shops in the US, Canada, Australia and Japan. Companies named in reports include Deere, SharkNinja and Cummins, with different levels of impact.
- The success rate was about 10%. A successful break-in took just under 23 hours on average.
- The apparent goal was lasting access to shopping and payment systems.
How the AI agents broke in
The agents didn’t use new magic. They used old, well-known weaknesses, just much faster and at much bigger scale:
- SQL injection and login bypasses on poorly written forms.
- Unsafe file uploads that let them place a “webshell”, a hidden script for running commands on the server.
- Outdated software, such as old framework components with known flaws.
- Exposed admin panels: when the main site was secure, agents looked for control panels left open to the internet.
- Leaked secrets: researchers found hundreds of exposed passwords, keys and test dashboards.
Why this matters for every website owner
Until now, attacking hundreds of sites needed a skilled team. AI agents let one person do it cheaply and around the clock. Small and mid-sized online stores, the kind that rarely have a security team, are exactly what these agents go after.
10 steps to protect your website now
- Update everything: your CMS (WordPress, Magento, etc.), themes, plugins and server software. Most attacks use known, already-fixed flaws.
- Hide or lock admin panels: restrict wp-admin, Magento admin, phpMyAdmin and hosting panels by IP, VPN or an extra password layer.
- Turn on two-factor login for every admin account.
- Restrict file uploads: allow only the file types you need and block script files from running in upload folders.
- Use a web application firewall (WAF): many hosts and CDNs offer free or low-cost options.
- Remove test and staging sites from the public internet, or protect them with a password.
- Never store keys or passwords in code or public files. Rotate any you suspect were exposed.
- Watch for new or changed files in your web folders. An unexpected PHP file in an uploads folder is a red flag.
- Review logs for unusual requests, especially repeated attempts on login pages and forms.
- Keep offline backups so you can restore a clean copy fast.
If you find a webshell: take the server offline, preserve logs, check for stolen data and other backdoors, and restore only from a clean backup after fixing the hole.
The bigger picture
Security companies are also using AI to defend: scanning code, spotting attacks and patching faster. Microsoft, for example, has added controls that let admins limit or shut down AI agents inside their systems. The race between AI attackers and AI defenders has started, and basic security hygiene is still the best protection.
Frequently asked questions
Can AI agents really hack websites on their own?
In this case, AI agents did most of the work with limited human involvement, using known weaknesses rather than new techniques.
Which websites are most at risk?
Sites running outdated software, exposed admin panels or weak upload forms, especially small online stores without dedicated security staff.
How much did the attack cost?
Researchers estimated about $5,000 in AI usage for the main campaign, or roughly $50 per successful break-in.
What is a webshell?
A small malicious script uploaded to a web server that lets an attacker run commands on it remotely.
Sources: Reuters investigation via Cybersecurity News and WinBuzzer; NorForge research; The Neuron. Figures are researcher estimates.
