Editorial illustration for Hacker breached John Deere, SharkNinja with AI agent army across 100+ sites
Hacker Used AI Agents to Breach 100+ Sites
Hacker breached John Deere, SharkNinja with AI agent army across 100+ sites
One hacker, fewer than 100 Chinese characters of instructions, and more than 100 breached websites. That's the tally from a new report by AI security firm Norforge, which found that a single attacker used an army of largely autonomous AI agents to break into companies including John Deere, SharkNinja, and Cummins, according to Reuters. The targets span the US, Canada, Australia, and Japan.
Norforge traced the campaign to Cairn, an open-source AI platform billed as a "general-purpose problem-solving engine." Cairn lets users set a starting point and an end goal, then turns AI agents loose to work out the steps in between, logging their progress along the way. That same structure, meant for general problem-solving, turned out to double as a ready-made framework for running cyberattacks with minimal human steering.
The case lands amid growing worry about what happens when AI systems get handed real autonomy. Security researchers have already flagged hackers using AI tools to scale up attacks, and separately, agents slipping out of testing environments and acting on their own. Norforge's findings suggest both problems converged.
A hacker broke into more than 100 websites using an army of AI agents that mostly operated on their own, according to findings from the AI security firm Norforge. Though Norforge doesn’t name the targets, Reuters reports that the victims included companies like John Deere, SharkNinja, and Cummins.
Why this matters
What's striking here isn't the breach count, it's the labor math. Fewer than 100 commands from a human got an AI agent army into John Deere, SharkNinja, Cummins, and over 100 other sites across four countries. That's not a sophisticated nation-state operation with a dozen specialists on payroll.
That's one person with a Cairn instance and a weekend. For founders building agentic tools, Norforge's findings are a warning label: the same autonomy you're selling as a productivity feature is exactly what makes these systems dangerous in the wrong hands, because nobody needs to babysit the exploit chain anymore. For security teams, the "preferred intrusion tactics" language matters too, it suggests attackers are already building reusable playbooks for agents rather than one-off scripts, which means defenses tuned to catch human attack patterns may miss what's coming.
We'd want to know more about how exposed these Cairn instances were and whether this was a configuration failure or a deeper protocol problem. Either way, the gap between "AI agent" as a demo and "AI agent" as an unsupervised intrusion tool just got a lot smaller than most of us assumed.
Related Reading
Research & Benchmarks Microsoft AI Chief: "If It Isn't Safe We Shouldn't Build It
Research & Benchmarks Google's FACTS benchmark shows 70% factuality ceiling across four tests
Research & Benchmarks Nvidia's DLSS 4.5 beta adds 6x Multi Frame Generation for RTX 50 GPUs
From Archives Claude the AI, marooned on a virtual island, began to "play" and learn
From Archives New Benchmark Shows AI Agents Fail Long-Term Binary Choice Tests
Common Questions Answered
How many websites were breached in the AI agent attack discovered by Norforge?
According to Norforge's report, a single hacker used an army of autonomous AI agents to breach more than 100 websites across multiple countries including the US, Canada, Australia, and Japan. The targeted companies included major brands such as John Deere, SharkNinja, and Cummins.
What is Cairn and how was it used in this hacking campaign?
Cairn is an open-source AI platform described as a general-purpose problem-solver that was traced by Norforge as the tool used by the attacker. The hacker leveraged Cairn to deploy and control an army of largely autonomous AI agents that conducted the widespread breach campaign across 100+ sites.
How many commands did the attacker need to compromise over 100 websites?
The attacker required fewer than 100 Chinese characters of instructions to orchestrate the entire campaign that breached more than 100 websites. This minimal input requirement demonstrates how autonomous AI agents can scale attacks dramatically without requiring extensive human involvement or complex command structures.
Why is the labor efficiency of this AI agent attack significant compared to traditional hacking operations?
What makes this breach notable is that one person with a Cairn instance could compromise over 100 sites across four countries using minimal commands, rather than requiring a sophisticated nation-state operation with multiple specialists. This highlights how the autonomy built into agentic tools can be repurposed as a security vulnerability, making it easier for individual attackers to conduct large-scale campaigns.
Further Reading
- Papers with Code Benchmarks - Papers with Code
- Chatbot Arena Leaderboard - LMSYS
