A Chinese-speaking hacker used autonomous artificial intelligence agents to break into websites linked to more than 100 companies, including agricultural equipment manufacturer Deere, home appliance company SharkNinja and engine maker Cummins, according to a cybersecurity investigation reported by Reuters on October 19, 2026.

The campaign targeted more than 1,200 websites, with the hacker delegating much of the work to AI agents rather than manually carrying out each step. Snorre Fagerland, a principal threat researcher at cybersecurity company NorForge, tracked the activity while investigating exposed instances of Cairn, an open-source AI problem-solving platform.

The researcher estimated that the operation consumed approximately $5,000 in AI usage, or around $50 per successful intrusion. Successful attacks typically involved about 23 hours of automated probing and exploitation.

The findings highlight how AI agents can help cybercriminals scale attacks by automating tasks that would otherwise require considerable time and technical effort.

However, the full impact remains unclear. Reuters could not establish how much data, if any, was extracted without authorisation from the affected organisations.

How the AI-Powered Attack Worked

According to Fagerland's investigation, the hacker used AI agents to investigate potential weaknesses in targeted websites and carry out intrusions with limited human direction.

The campaign appears to have involved tools including Cairn, DeepSeek and Anthropic's Claude Code. The hacker reportedly supplied brief instructions describing the desired attack approach, leaving the agents to perform much of the subsequent work.

The investigation began after Fagerland discovered exposed Cairn instances being used for malicious activity. Access to the activity logs allowed him to observe how operators deployed AI tools against online targets.

The campaign began around September 25 and targeted e-commerce websites in the United States, Canada, Australia and Japan. By the end of the observed period, the agents had attempted to compromise more than 1,200 websites.

The important development is not simply that a hacker used AI to write code or research vulnerabilities. The reported agents were tasked with carrying out extended sequences of technical work, allowing one operator to pursue many targets in parallel.

This does not mean AI eliminates the need for human direction, suitable tools or exploitable weaknesses. It does show how automation can reduce the amount of direct effort required for a sustained attack.

Deere, SharkNinja and Cummins Among the Identified Companies

Reuters identified several companies whose websites appeared in the attack activity. Their responses indicate that the incidents differed in scope and that a successful intrusion does not automatically mean sensitive corporate data was stolen.

  • Deere: The agricultural equipment manufacturer said it had taken necessary remediation measures and had no indication that sensitive information was accessed.

  • SharkNinja: The home appliance company said the affected site was a standalone website containing product-support information and that the threat had been contained.

  • Cummins: The engine manufacturer said it was investigating a report of a potential cyberattack involving a website operated by one of its third-party vendors.

Two smaller businesses also acknowledged being compromised, according to Reuters. Other organisations listed in the research did not respond to requests for comment.

The distinctions matter. A website breach, unauthorised access to a third-party system and confirmed theft of sensitive data are different outcomes. The investigation has not established the full extent of data extraction across the reported victims.

Why AI Agents Change the Economics of Cybercrime

Traditional cyberattacks can require people with specialised skills to research targets, identify vulnerabilities, develop exploits and adapt when an attempted intrusion fails.

AI agents can automate parts of these processes and keep working through multi-step tasks with less continuous human supervision. A malicious operator can therefore direct the same tools towards a larger number of targets without personally performing every technical action.

The estimated $5,000 cost of the reported campaign illustrates the potential economics. If a relatively small expenditure can support attempts against more than 1,200 websites, the cost of testing a large number of targets may become less of a barrier for criminals.

But the estimated $50 per successful intrusion should not be interpreted as a universal cost for hacking a company. It is a researcher’s estimate for this particular operation, based on its reported AI usage. It does not include every possible cost of developing tools, obtaining access or monetising stolen information.

The incident also does not prove that AI can bypass every security control. The attacks depended on weaknesses that the agents could identify and exploit.

What Businesses Can Do to Reduce the Risk

The case reinforces the importance of treating AI-assisted attacks as a practical cybersecurity concern rather than a distant possibility.

Businesses can reduce exposure by strengthening several areas of their security programmes:

1. Patch exposed systems promptly. Keep websites, content management systems, software libraries and internet-facing services updated. Prioritise known exploited vulnerabilities and systems exposed directly to the internet.

2. Monitor unusual activity. Look for repeated probing, unusual request patterns, suspicious login attempts and unexpected activity across web applications. Automated agents can conduct long-running attacks, so detection should not depend on a human attacker making obvious mistakes.

3. Secure third-party services. The Cummins response illustrates why vendor security matters. Organisations should know which external providers operate their websites and systems, what access those providers have, and how incidents will be reported.

4. Limit access and privileges. Apply least-privilege access, multifactor authentication and appropriate separation between systems. If one service is compromised, attackers should not automatically gain access to an entire corporate environment.

5. Prepare incident-response plans. Businesses should know how to isolate affected systems, preserve evidence, rotate compromised credentials and investigate possible data exposure. Recovery should include identifying and fixing the original weakness, not simply taking a website offline temporarily.

6. Use security testing responsibly. Organisations can apply automated security tools to systems they own or are authorised to test, with clear scope and safeguards to prevent disruption.

The reported campaign shows how AI agents may make cyberattacks more scalable by allowing one operator to automate hours of reconnaissance and exploitation across many targets.

The immediate concern for businesses is not that AI makes every system vulnerable. It is that attackers may be able to test more systems, more persistently and at a lower marginal cost than before.

The reported breaches also demonstrate why cybersecurity needs to cover third-party websites and smaller internet-facing services, not just a company's most prominent systems. A neglected support portal or vendor-operated site can still create risk.

The full damage from this campaign remains uncertain, and the estimated cost should be treated as specific to this case. Even so, businesses should prepare for attackers who can use AI to accelerate familiar techniques.