One Hacker Used AI Agents to Breach More Than 100 Websites: What We Know

Cybersecurity experts are warning about a new phase of cybercrime after a hacker reportedly used artificial intelligence agents to compromise websites belonging to more than 100 companies. Instead of carrying out every step manually, the attacker relied on an autonomous AI framework to investigate potential weaknesses and conduct attacks with limited human supervision.

The campaign reportedly targeted more than 1,200 websites across the United States, Canada, Australia, and Japan. Companies named in reporting about the incident include John Deere, SharkNinja, and Cummins.

The discovery highlights how AI-powered automation could allow individual cybercriminals to conduct attacks that previously required substantially more time and technical effort.

How the AI-Powered Hacking Campaign Worked

According to cybersecurity researcher Snorre Fagerland of NorForge, the attacker used Cairn, an open-source, multi-agent AI framework, to coordinate much of the operation.

AI agents are software systems that can pursue objectives, complete intermediate tasks, and adjust their approach based on results. Unlike a simple script that follows a fixed sequence of instructions, an AI agent can attempt different approaches when its initial efforts fail.

In this case, the agents reportedly investigated targeted websites, searched for security weaknesses, and continued working through extended sessions with little direct human intervention.

The attacker provided relatively brief instructions, leaving the AI systems to perform much of the subsequent work.

Researchers observed that successful intrusions typically took around 23 hours of investigation and experimentation. The campaign demonstrates how autonomous systems can potentially sustain complex tasks for long periods without requiring a human operator to guide every decision.

However, AI did not eliminate the human role entirely. The attacker still selected targets and established the objectives for the campaign.

John Deere, SharkNinja, and Cummins Among the Named Companies

The incident attracted particular attention because several recognizable companies were identified among the affected organizations.

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

SharkNinja: The home appliance company said the incident affected a standalone website used for product support and that the threat had been contained.

Cummins: The engine manufacturer said it was investigating a reported potential cyberattack involving a website operated by a third-party vendor.

These responses are important because a reported website compromise does not automatically mean that customer records, financial information, or internal corporate systems were stolen.

The available reporting did not establish that sensitive data had been extracted from these three companies. The precise impact of the wider campaign remains uncertain.

How Much Did the AI Attacks Cost?

One striking aspect of the campaign was its reported operating cost.

Fagerland estimated that the operation consumed approximately $5,000 worth of AI processing tokens. Based on the reported scale of successful intrusions, the estimated cost was around $50 per successful breach.

These figures are estimates rather than independently verified costs for every compromised website.

Nevertheless, they illustrate why cybersecurity professionals are paying attention to autonomous AI tools. If attackers can automate time-consuming investigation and testing, they may be able to pursue more targets without hiring a large team of specialists.

Lower operational costs could also make sophisticated attack campaigns accessible to a broader range of criminals.

The financial calculation does not include the full costs of operating criminal infrastructure, obtaining access, managing stolen information, or avoiding detection.

Why Autonomous AI Agents Are Different From Traditional Hacking Tools

Traditional cybersecurity tools often automate specific activities, such as checking systems for known vulnerabilities or identifying exposed services.

AI agents can go further by selecting intermediate tasks, evaluating results, and changing their approach when an earlier attempt fails.

According to the reporting, the agents involved in this campaign sometimes continued investigating after their initial attempts did not reveal an obvious weakness. This ability to persist and adapt was one reason the case concerned researchers.

The distinction matters because defenders cannot rely exclusively on identifying a familiar attack pattern. Security systems may also need to detect unusual behavior, unexpected access attempts, and suspicious sequences of activity.

However, AI does not make every attack successful, and conventional security weaknesses remain an important part of the problem. The technology can help attackers automate existing techniques, but its effectiveness depends on the target environment and the vulnerabilities available.

The Role of Cairn in the Campaign

Cairn is an open-source platform designed around a multi-agent architecture. Its general-purpose approach allows users to coordinate AI-driven tasks, but the same flexibility can be misused.

Fagerland reportedly discovered more than two dozen exposed Cairn instances associated with malicious activity while investigating the broader threat landscape.

In the campaign under discussion, the exposed infrastructure gave the researcher an opportunity to observe activity and examine how the agents were being used.

Cairn's creator, who uses the online handle le1xia0, said the platform was not designed as a hacking tool and that harmful use was not the intended purpose.

This distinction is relevant to the wider debate about open-source AI. A technology developed for legitimate automation can potentially be repurposed for harmful activities, creating challenges for developers, platform operators, and cybersecurity teams.

The incident also demonstrates why AI orchestration systems must be secured carefully, particularly when they can interact with external services or execute actions with significant privileges.

Why AI-Driven Cybercrime Is Becoming a Bigger Concern

The campaign is part of a broader pattern of reported incidents involving AI-assisted cyberattacks.

Security researchers have documented cases in which criminals used AI tools to accelerate reconnaissance, investigate weaknesses, and automate parts of their operations.

In September 2026, another report described an AI-assisted attack that resulted in breaches at multiple companies and the theft of hundreds of thousands of payment-card records. These incidents illustrate the potential consequences when automation is combined with weak security controls.

AI agents may make it easier for attackers to work continuously, investigate more targets, and adapt their behavior. At the same time, defenders can use AI to analyze security alerts, identify suspicious patterns, and accelerate incident response.

The technology therefore creates opportunities for both attackers and defenders, although the balance depends on how effectively organizations deploy and govern their systems.

How Businesses Can Protect Their Websites

Organizations should not assume that traditional security practices are no longer effective. Instead, the rise of autonomous attacks makes consistent security measures even more important.

Businesses can reduce their exposure by taking several practical steps:

• Patch known vulnerabilities: Keep website software, servers, frameworks, and dependencies updated.
• Secure public-facing systems: Review internet-accessible applications, administrative interfaces, and third-party services.
• Use strong access controls: Require multi-factor authentication where appropriate and restrict administrative permissions.
• Monitor unusual activity: Investigate unexpected login attempts, abnormal requests, and suspicious behavior across applications.
• Protect AI infrastructure: Avoid exposing agent-control dashboards publicly and isolate automated systems from sensitive production environments.
• Limit automated permissions: Require human approval for high-risk actions and restrict what AI agents can access or modify.
• Prepare an incident-response plan: Establish procedures for isolating affected systems, preserving evidence, and notifying the appropriate teams.

Companies should also evaluate the security of vendors that operate websites or services on their behalf. As the Cummins statement illustrates, third-party infrastructure can be relevant to an investigation even when a company's primary systems are not directly identified as the point of compromise.

Could AI Agents Make Hacking More Accessible?

The case raises concerns that cybercriminals may no longer need to perform every technical task themselves. With suitable tools and objectives, an individual may be able to delegate parts of a complex operation to AI systems.

That does not mean expertise has become unnecessary or that anyone can automatically breach a well-protected organization. Attackers still face technical barriers, security monitoring, access restrictions, and the possibility that their efforts will fail.

Nevertheless, the ability to automate investigation and repeat attempts at scale could change the economics of cybercrime.

For cybersecurity teams, the challenge is to identify threats earlier, secure exposed systems, and ensure that automated tools cannot operate without appropriate restrictions.

Final Thoughts

The reported campaign involving Cairn illustrates how autonomous AI agents can be misused to conduct large-scale cyberattacks with relatively limited human involvement.

More than 1,200 websites were reportedly targeted, and over 100 company websites were said to have been compromised. John Deere, SharkNinja, and Cummins were among the named organizations, although the confirmed impact differed between companies.

The estimated $5,000 in AI processing costs also highlights why security researchers are concerned about the growing accessibility of automated attacks.

AI agents are not a replacement for every traditional hacking technique, but they can increase the speed, persistence, and scale of certain operations. Businesses should respond by strengthening basic security, protecting AI infrastructure, monitoring suspicious activity, and keeping human oversight over high-risk automated actions.

As AI technology advances, cybersecurity will increasingly depend