One person. Fewer than 100 characters of instruction. About $5,000 in AI tokens. That, according to a Reuters investigation published on Monday, October 19, was enough to break into the websites of more than 100 companies — including farm equipment giant Deere, home appliance maker SharkNinja and engine manufacturer Cummins. The hacker did not sit at a keyboard picking locks one by one. He handed almost the entire job to autonomous AI agents, and then mostly watched them work, unsupervised, for up to 23 hours at a stretch. Security researchers who observed the campaign say it is the clearest proof yet that the age of the lone, AI-powered hacking operation is not coming — it has already arrived.
The findings come from Snorre Fagerland, a principal threat researcher at the cybersecurity firm NorForge who has spent nearly three decades in the field. Fagerland told Reuters he had expected hacking agents to still need a human looking over their shoulder. This campaign changed his mind. "It's a new world now," he said.
Key Takeaways
- On October 19, 2026, Reuters reported research from NorForge showing one Chinese-speaking hacker used autonomous AI agents to attempt break-ins at more than 1,200 websites and breach more than 100 companies since September 25.
- Named victims include Deere, SharkNinja and Cummins, across the United States, Canada, Australia and Japan. Each successful breach took about 23 hours and cost roughly $50 in AI tokens — about $5,000 in total.
- The agents ran on Cairn, an open-source AI problem-solving platform, in a setup NorForge reads as Anthropic's Claude Code calling DeepSeek's low-cost V4.1 Flash model.
- Deere says there is no indication sensitive information was accessed, SharkNinja says a standalone support site was contained, and Cummins says it is investigating a vendor-operated website.
- For everyday users, the real near-term risk is follow-on phishing: fake "breach notice" emails that steal passwords and payment details by pretending to help.
How One Hacker Ran a 1,200-Site Hacking Spree With AI Agents
The campaign began around September 25 and targeted e-commerce websites in the United States, Canada, Australia and Japan — sites pre-filtered, NorForge says, by traffic volume and tested to confirm that orders could actually be placed on them. The operator's entire instruction set fit in fewer than 100 Chinese characters: in effect, find a way to run code on each site, then plant an "uploader" web shell — a small back door that lets an attacker upload files later and monetize the access.
From there, the agents worked on their own. They probed targets for weaknesses, tried entry points, backed off when a route failed and hunted for another. Where they succeeded, the full run from first scan to working access typically took about 23 hours. By late last week, the agents had attempted more than 1,200 sites and succeeded against more than 100 of them. Fagerland put the total token bill at roughly $5,000 — about $50 per successful hack, less than a single hour of a junior penetration tester's time, delivered by a workforce that never sleeps and never gets bored re-checking the same login page.
The tool underneath it all was Cairn, an open-source platform that lets a user define a starting point and a goal for a group of AI agents, then tracks their progress on a shared board. Fagerland found the operation almost by accident: while scanning the open web from around October 2 for carelessly exposed Cairn deployments, he found roughly two dozen instances being used for malicious purposes — their control panels left open enough for him to watch the work unfold.
The Method Was Not "Script Kiddie Stuff"
What unsettled researchers most was not the volume. It was the problem-solving. In one case Fagerland described, the agents found no opening in a large company's public-facing application, so they combed through digital records to identify related infrastructure, located an exposed control panel, spotted a subtle flaw in how it handled data and, after hours of trial and error, used it to make the server run malicious commands.
"It's not straightforward to come up with solutions like that," Fagerland told Reuters. "It's not script kiddie stuff." That distinction matters for American businesses of every size. Until now, a mid-sized online store could reasonably assume it was beneath the notice of a skilled human attacker. At $50 a breach, that assumption no longer holds — automated agents can profitably aim at targets that were never worth a human's afternoon.
What the Named Companies Are Saying
Deere said it carried out "necessary remediation actions" and found "no indication that sensitive information was accessed." SharkNinja said the attack struck a standalone website holding only product support information, and that the threat has been contained. Cummins said it is "investigating the report of a potential cyber-attack on a website operated by one of our third-party vendors." Two smaller businesses also confirmed to researchers that they had been compromised. Importantly, NorForge could not determine how much data, if any, was actually taken in the intrusions — so no report should claim customer data was stolen from these three companies, because that has not been established.
The Chinese-speaking operator was the most productive of several actors Fagerland observed abusing exposed Cairn panels. Others were less restrained: one made off with a large cache of credit card data, and another aimed agents at a cryptocurrency exchange. That spread is exactly why researchers are alarmed — the same playbook is already in several hands at once.
This Is Part of a Wider AI Hacking Wave, Not a One-Off
The NorForge disclosure lands in a busy, uncomfortable stretch for AI security. In August, the Israeli firm Gambit Security reported Russian-speaking extortionists using an AI coding tool to break into seven organizations. In September, the same firm documented an agent-driven campaign that compromised 25 companies and yielded at least 600,000 stolen credit card records. Last week, cyber intelligence company Bitsight said chatter about deploying autonomous agents was climbing sharply on criminal forums. And it follows the pattern we covered in our report on the OpenCode breach that exposed OpenAI, Anthropic and Google API keys — AI tooling is creating both new targets and new weapons at the same time.
Cairn's creator, who uses the handle "le1xia0," pushed back on the framing. In a recent post on X, the developer described the platform as "an innovative type of multi-agent architecture" that "was not designed for hacking attacks," adding that "misuse is a human decision, not a feature of the engine." NorForge says it is working to get the malicious instances taken down.
Who Is Affected — and Who Should Actually Worry?
If you bought a SharkNinja appliance, a Cummins-powered product or Deere equipment and registered it online, do not panic — and do not ignore the story either. The confirmed compromises were websites, including a support site and a vendor-run site, not an announced theft of customer databases. The realistic risks for everyday users in the US are second-order ones:
- Phishing that rides the headline. Scammers read breach news too. Expect emails and texts claiming to be from a named brand, urging you to "verify your account" after the hack.
- Smaller stores, bigger exposure. Most of the 100-plus victims were not household names. If a small online shop you use announces an incident in the coming weeks, treat its password as burned and change it everywhere you reused it.
- Stored payment details. Other operators in the same ecosystem did steal credit card data. Check statements on cards saved with smaller e-commerce sites.
This is also a good moment to tighten how much personal material sits in cloud accounts tied to shopping logins. Keeping ID scans, receipts and private documents in a local, on-device vault — rather than in an email inbox an attacker can search after a password reset — reduces what any single breach can reach. That is the idea behind Pro Calculator Secure Vault, which stores photos, videos, audio, documents and private notes locally on your device, with no data collected or shared with third parties.
Your 15-Minute Safety Checklist After This News
- Change the password on any SharkNinja, Cummins or Deere account you own — and anywhere you reused the same password.
- Turn on two-factor authentication for shopping and brand accounts, preferably with an authenticator app rather than text codes.
- Review card statements for the last 60 days for cards saved on smaller online stores, and dispute anything you do not recognize.
- Do not click links in "breach notification" messages. Type the brand's real web address into your browser yourself.
- Move scans of IDs, tax papers and private photos off your email and gallery into an offline vault on your phone.
- If you run a website, patch it, close exposed admin and AI-agent control panels, and audit third-party vendor access this week.
Warning: Fake "Breach Help" Is the Scam That Follows Every Big Hack Story
Within days of coverage like this, inboxes fill with messages that borrow a real company's logo and a frightening subject line: "Your data was found in the Deere breach — verify now." None of the three named companies has announced a customer notification program from this incident, and legitimate breach notices never ask you to enter a full card number or pay a fee to "protect" your account. If a message pressures you to act within hours, that urgency is the tell. Go directly to the company's official site or app, check for a statement there, and only then decide whether you need to do anything at all.
Frequently Asked Questions
Which companies were breached by the AI agents?
The named, confirmed companies are Deere, SharkNinja and Cummins, out of more than 100 compromised websites overall. Most victims have not been publicly named, and two smaller businesses also confirmed compromises to researchers.
Was customer data stolen from Deere, SharkNinja or Cummins?
That has not been established. Deere says there is no indication sensitive information was accessed. SharkNinja says only a standalone support site was hit and the threat was contained. Cummins is still investigating a vendor-operated website. NorForge could not determine how much data was removed in the intrusions.
How much did the hacking campaign cost to run?
NorForge estimates roughly $5,000 in AI model tokens in total — about $50 for each successful breach — with each success taking around 23 hours of unsupervised agent work.
What is Cairn, and is it a hacking tool?
Cairn is an open-source platform that coordinates AI agents toward a defined goal and tracks their progress. Its creator says it was built as a general multi-agent system and was never designed as a hacking tool; in this campaign, an attacker repurposed it to find and exploit website weaknesses.
What should I do if I have an account with one of these brands?
Change that password, make sure it is not reused anywhere else, enable two-factor authentication, and watch the payment method on file. Keep private documents and photos in a local vault such as Pro Calculator Secure Vault — with sign-in protection, an intruder selfie feature and an incognito private browser, plus 8 smart calculators including standard, floating, BMI, age, currency, tip, GST & VAT and loan EMI tools — so a compromised web account cannot open them.
The Bottom Line
The most important number in this story is not 100 companies or 1,200 websites. It is $50 — the approximate cost, in AI tokens, of one successful corporate break-in run by software that needed fewer than 100 characters of human instruction. Defenders, insurers and regulators are only beginning to price that reality in. Expect more disclosures like NorForge's in the coming months, tougher questions for companies about exposed AI tooling and vendor websites, and a fast-growing market for tools that watch agents the way we once watched logins. Until then, unique passwords, two-factor authentication and keeping your most private files offline remain the cheapest defense an ordinary user can buy. Pro Calculator Secure Vault is built by ZAS — you can see more of our apps on our developer page on Google Play.
No comments:
Post a Comment