What happened
NorForge, an AI security firm, published a report on October 19, 2026 about Cairn, an open-source platform that coordinates AI agents toward a goal. Cairn can be used for legitimate penetration testing. NorForge found more than 20 Cairn control panels left open on the internet that it assessed as likely used by criminals, which let its researcher watch their work. Reuters reported that the most active operator, a Chinese-speaking hacker, started on September 25 and attempted more than 1,200 e-commerce sites in the US, Canada, Australia, and Japan. Reuters says the hacker broke into websites of more than 100 companies, including Deere, SharkNinja, and Cummins. The instructions were under 100 Chinese characters. Successful intrusions typically took around 23 hours of trial and error. Researcher Snorre Fagerland estimated about $5,000 in AI usage, roughly $50 per successful break-in. Reuters says the activity appeared to use a mix of DeepSeek and Anthropic's Claude Code. The companies responded in different ways. Deere told Reuters it took remediation steps and has no indication that sensitive information was accessed. SharkNinja said a standalone site with only product support information was hit and the threat was contained. Cummins said it is investigating a possible attack on a website run by a third-party vendor. Reuters could not tell how much data, if any, was taken. NorForge also reports that the three clusters it tracked succeeded in roughly 10% of attempts, and that it is working to have the panels taken down.
Why it matters for business owners
The story is not that a big company got hacked. It is how little the attacker had to do. A short instruction, a list of targets, and a few thousand dollars of AI usage produced attempts against more than a thousand sites. At that cost, a business does not need to be famous to be worth trying. Mid-size companies with an online store, a customer portal, or a forgotten marketing site are inside the same net. The named victims also show where the weak spot often is. SharkNinja's affected site was a standalone support site. Cummins pointed to a site run by a vendor. These are the sites that nobody on the leadership team thinks about. There is a second point. NorForge says the open panels exposed the targets' secrets, and that a scan found hundreds of credentials, 11% of which were verified valid. AI tooling left open on the internet leaks what it touches. That applies to legitimate teams too, including any consultant running tests on your behalf.
What owners should not misunderstand
Do not read this as proof that AI makes every attack succeed. By NorForge's estimate, roughly one attempt in ten worked. The change is volume and cost, not magic. Do not assume the headline numbers are precise. The $5,000 and $50 figures are one researcher's estimates, and another outlet reports NorForge's own average at a lower cost per task with a large margin of error. Treat them as scale, not as a price list. Do not assume the models named are the problem. Reuters says the campaign appeared to use DeepSeek and Claude Code, and Cairn's author says the platform was not designed for hacking. Blaming one tool misses the point. The same capability is available from many sources. Do not assume that a breach of a minor site is minor. Reuters could not establish what was taken, and NorForge describes the aim of the largest cluster as lasting access to payment and online store infrastructure for later use. And do not buy something new in a panic. The fixes here are mostly basic.
The operational lesson
You cannot protect what you cannot list. Most businesses run more web properties than anyone can name in a meeting: the main site, a store, a booking page, a support portal, a campaign microsite, a vendor-hosted login, an old admin panel. Each is a door. The attacker's agents are patient and do not get bored of the boring ones. The lesson is ownership. Every site and portal needs a named person, a known host, a patch routine, and an account on file for who can sign in. When a vendor runs it, the vendor relationship needs the same answers in writing.
What a serious business should do next
Make a list of every site, portal, and admin panel tied to your business, including those a vendor, agency, or former employee set up. Check your domain registrar and DNS records, because they show what is actually live. For each one, write down the owner, the host, the platform or plugins it runs, and when it was last updated. Shut down anything unused. Patch the software behind your online store and any admin panels. The attacks described targeted ordinary weaknesses such as unsafe file uploads, login flaws, and exposed control panels. Remove public access to admin panels where you can, or require a second sign-in step. Ask each vendor that runs a site or portal for you how they patch it, who monitors it, and how fast they would tell you about an incident. Get the answer in writing. Put one person on daily review of sign-in and admin alerts for the store and payment pages. If you have no logs, ask your host how to turn them on. If anyone tests your systems with AI-driven tools, ask how the results are stored and who can reach the dashboard. NorForge found that open test panels exposed targets' data. Rotate any password or key that has sat in an old system for years. Then confirm someone knows what to do on the day an alert arrives. If you have a real incident, bring in a qualified security professional. This post is not security or legal advice.
The Atlacis view
Atlacis helps owners slow down before an AI decision, map the workflow, and see what each tool touches. The same discipline applies to the other side of this story. When attackers use AI to test everything cheaply, the businesses that fare better are the ones that know what they run and who answers for it. If you are not sure which AI tools, vendors, and web systems your team depends on, mapping them is a sensible first step.
The short version
- NorForge and Reuters report that one hacker's AI agents attempted more than 1,200 websites and got into sites of more than 100 companies, including Deere, SharkNinja, and Cummins.
- The estimated cost was about $5,000 of AI usage, roughly $50 per successful break-in. These are researcher estimates, and roughly one attempt in ten worked.
- The weak points named were a standalone support site and a vendor-run site. Forgotten and outsourced sites are where cheap automated probing tends to land.
- List every site, portal, and admin panel your business depends on, name an owner for each, and shut down what you do not use.
- Ask vendors who run sites for you how they patch, monitor, and report incidents, and get it in writing. Involve a security professional for any real incident.
Where ATLACIS can help
Sources
- NorForge: Inside Cairn, open windows into massive AI-driven cyber operations (October 19, 2026)
- Reuters via Rappler: How one hacker used AI agents and a few thousand dollars to break into scores of sites (October 20, 2026)
- The Verge: A hacker broke into John Deere and SharkNinja with an army of AI agents (October 19, 2026)
- SecurityWeek: AI platform Cairn abused to hack over 100 organizations (October 20, 2026)