AI-powered cyberattack uses autonomous agents to breach 100+ sites

A hacker used an open‑source AI platform called Cairn to launch an army of autonomous agents that broke into more than 100 websites, according to The Verge. The campaign hit well‑known companies such as John Deere, SharkNinja and Cummins, and the attacks were completed in roughly 23 hours each.

How the attack was carried out

Researchers at the AI security firm Norforge discovered the activity while scanning public instances of Cairn. The platform lets users define a start and end point for AI agents, which then document each step they take. While the tool is marketed as a general‑purpose problem‑solving engine, Norforge found that a malicious actor had turned it into a coordinated hacking suite. The attacker set "preferred intrusion tactics" in fewer than 100 Chinese characters, and the AI agents executed those tactics without further human input.

Why it matters for small businesses and nonprofits

Even though the headline victims are large manufacturers, the technique can be applied to any web‑based system that runs on standard software stacks. An open‑source tool is freely available, and the only barrier is the attacker’s willingness to repurpose it. For organizations that lack dedicated security teams, the prospect of an autonomous AI agent probing their site can be unsettling. The breach at SharkNinja, for example, affected a standalone support site, showing that even ancillary domains are at risk.

What the affected companies are saying

John Deere’s spokesperson told Reuters there is “no indication that sensitive information was accessed,” while Cummins said it is “investigating the report of a potential cyber‑attack on a website operated by one of our third‑party vendors.” SharkNinja confirmed the breach was limited to a support site and that it has contained the threat. Norforge is working with hosting providers to take down the malicious Cairn instances, and the platform’s creator emphasized that the tool itself was never intended as a hacking device, adding that “misuse is a human decision, not a feature of the engine.”

Practical steps to reduce your risk

• Inventory all public‑facing web assets, Know every domain, subdomain and third‑party site that your organization runs.
• Apply the principle of least privilege, Limit what each web service account can do; remove unnecessary admin rights.
• Keep software up to date, Patch content management systems, plugins and server packages as soon as updates are released.
• Enable multi‑factor authentication, Require MFA for any administrative login, especially for CMS and hosting dashboards.
• Monitor for unusual activity, Use logging and alerting tools to spot repeated login attempts, unfamiliar IP addresses, or sudden changes to files.
• Restrict outbound connections, Prevent web servers from reaching unnecessary external services that could be used to fetch malicious code.
• Conduct regular security reviews, Run vulnerability scans and consider a third‑party penetration test at least annually.

Looking ahead: AI tools and security testing

The incident underscores a broader trend: as AI tools become more capable, they can be double‑used for both defense and offense. Security teams are beginning to experiment with AI‑driven testing to find weaknesses faster, but attackers can flip the same capabilities to scale their exploits. Organizations should stay informed about the tools their industry relies on and consider integrating AI‑aware controls into their security policies.

What you can do right now

1. Review your web‑hosting contracts, Ask providers what monitoring they perform for rogue AI activity.
2. Audit third‑party vendors, Confirm that any external partners follow strong security practices and have incident‑response plans.
3. Train staff on phishing and social engineering, Many AI‑assisted attacks start with credential theft; awareness reduces that risk.
4. Document a response plan, Outline steps for containment, communication and recovery should an AI‑powered breach occur.

By taking these steps, small businesses, nonprofits and public‑sector teams can make it harder for autonomous AI agents to gain a foothold, even if the tools they use were originally designed for legitimate purposes.

For a plain‑English AI roadmap tailored to your organization, visit our AI services page.