Artificial intelligence is changing the cybersecurity battlefield. A new investigation has uncovered a hacking campaign in which AI agents were used to probe more than 1,200 websites and reportedly break into systems belonging to over 100 companies, with very little direct work from the human operator.
The investigation, reported by Reuters on October 19, 2026, describes how a cybercriminal used autonomous AI agents to search for vulnerabilities, investigate infrastructure and attempt to gain access to business websites.
The affected companies reportedly included agricultural equipment manufacturer Deere, home appliance company SharkNinja and engine manufacturer Cummins. The incident highlights a growing concern for cybersecurity professionals: attackers may no longer need large teams of skilled hackers to conduct complex campaigns.
How the AI-Powered Hacking Campaign Worked
According to cybersecurity researcher Snorre Fagerland of NorForge, the campaign involved an open-source AI problem-solving platform called Cairn.
The attacker gave AI agents relatively short instructions and allowed them to work through potential targets with limited human intervention. Rather than manually checking every website, the agents could investigate weaknesses, follow leads and try different approaches over extended periods.
The campaign began in late September, with the agents targeting e-commerce websites in countries including the United States, Canada, Australia and Japan.
By the time researchers examined the activity, the agents had reportedly attempted to break into more than 1,200 websites.
The important distinction is that these were attempted intrusions, not 1,200 confirmed successful hacks. NorForge’s investigation indicated that websites belonging to more than 100 companies were compromised, although the amount of information accessed or stolen remains uncertain.
AI Agents Can Do More Than Scan for Vulnerabilities
Traditional automated security tools have long been used to identify exposed services, outdated software and common vulnerabilities. The difference with AI agents is their ability to investigate findings and adapt their next steps.
In one example described in the reporting, agents failed to find a useful weakness in a company’s public-facing application. They then investigated related infrastructure, located another publicly accessible control panel and explored a weakness in how that system processed data.
After extended attempts, the agents reportedly found a way to make a server execute malicious commands.
This kind of multi-step investigation can require experience and patience when performed manually. AI agents can potentially repeat the process across many targets without requiring a human to direct every action.
However, this does not mean AI can automatically break into every website. The campaign depended on weaknesses in real systems, and many attempted attacks do not succeed.
How Much Did the Attacks Cost?
One of the most concerning aspects of the investigation was the estimated cost.
Fagerland estimated that the campaign consumed roughly $5,000 worth of AI usage, equivalent to approximately $50 per successful intrusion, according to Reuters.
The figures are estimates rather than independently verified accounting records. Actual costs can vary depending on the models used, the number of failed attempts and the complexity of each target.
Even with that uncertainty, the investigation illustrates how relatively inexpensive AI services could allow one attacker to attempt operations at a scale that would otherwise require considerably more time and effort.
Lower costs do not make successful attacks inevitable, but they can make repeated attempts more practical for criminals.
Which Companies Were Affected?
Reuters identified several companies associated with the campaign.
Deere: The agricultural equipment manufacturer said it had taken necessary remediation measures and had no indication that sensitive information had been accessed.
SharkNinja: The home appliance company said the affected website was a standalone site 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. The full scope of the campaign and any resulting data exposure remained unclear in the reporting.
The incident is a reminder that attackers do not always need to compromise a company’s main website. Third-party vendors, support portals and connected infrastructure can also create security risks.
The Exposed AI Control Panels Created Another Security Risk
Researchers discovered the activity by looking for publicly exposed instances of Cairn.
These control panels provided visibility into the agents’ projects and, in some cases, information about their targets, findings and collected credentials or keys.
NorForge reportedly identified hundreds of secrets in the exposed data, with some still appearing to be valid. The researchers also found apparently legitimate penetration-testing operations whose exposed panels risked revealing client information.
This created a second problem: the infrastructure used to coordinate the attacks could itself expose sensitive information.
The findings underline a basic security principle. Tools designed for legitimate testing can become dangerous when deployed against systems without authorization or operated with inadequate access controls.
Cairn’s creator told Reuters that the platform was not designed for criminal hacking and that harmful use was not the intention.
What This Means for Website Owners
AI-assisted attacks make basic website security and monitoring even more important. Website owners should not assume that being a small business makes them an unlikely target.
Here are several practical steps that can reduce exposure:
- Keep software updated: Apply security patches to WordPress, plugins, themes, server software and other internet-facing applications.
- Remove unnecessary services: Disable old test environments, unused control panels and abandoned subdomains.
- Protect administrator accounts: Use multifactor authentication and strong, unique credentials.
- Review file uploads: Make sure upload functionality is restricted, validated and unable to execute dangerous files.
- Secure third-party services: Check the security of hosting providers, vendors, support portals and integrations.
- Monitor for suspicious activity: Review authentication logs, unexpected administrator accounts, unusual file changes and unexplained server processes.
- Protect secrets: Never expose API keys, access tokens, database credentials or other sensitive information in public dashboards or logs.
For WordPress site owners, keeping plugins and themes updated and removing abandoned extensions are particularly important first steps. A security plugin can help with monitoring, but it cannot replace proper server configuration, backups and timely patching.
Is AI Making Hacking Easier for Everyone?
The investigation suggests that AI can reduce the amount of manual work required to conduct certain cyberattacks. A single operator may be able to launch many investigations simultaneously and let agents pursue possible weaknesses.
But AI is not a magic key. Attackers still depend on exploitable vulnerabilities, misconfigurations, stolen credentials or other weaknesses. Agents can make mistakes, waste resources and produce misleading findings.
The same technology can also help defenders analyse logs, detect suspicious behaviour, investigate vulnerabilities and respond to incidents more quickly.
The challenge is ensuring that defensive capabilities keep pace with the growing accessibility of automated attack tools.
The Future of Cybersecurity Is Becoming More Automated
The most important lesson from this incident is not simply that AI can be used for hacking. Automated security testing has existed for years.
What makes this campaign significant is the reported ability of AI agents to investigate targets, adapt their approach and pursue multi-step intrusions with limited human involvement.
For businesses, that means security cannot depend on attackers being too slow or too expensive to reach them. Systems need to be patched, unnecessary access removed and suspicious activity detected before a small weakness becomes a serious incident.
AI is giving defenders new tools, but it is also giving criminals new ways to scale their operations.
Could an AI-powered attacker find a weakness in your website before you do?
As autonomous agents become more capable, that question is becoming increasingly relevant for businesses, website administrators and anyone responsible for an internet-connected system.
Sources: Reuters investigation and The Hacker News coverage. Details about the number of attempted attacks, affected companies, costs and data exposure are attributed to the reporting and research described in those sources.
