Chinese-speaking hackers have largely outsourced a maelstrom of malicious activity against numerous e-commerce websites to the AI-driven penetration testing tool Cairn, with researchers finding more than 20 publicly exposed dashboards engaged in highly automated activity bent on stealing payments.
The findings offer a rare look inside AI-assisted cyber operations, exposing not just reconnaissance activity but workflows designed to turn vulnerability discovery into working access to victim infrastructure. Importantly, the campaign emphasizes that AI is significantly broadening the target base for attackers, and organizations need to start changing up their defenses now to meet the threat.
The attackers targeted 100 sites across various global regions, including the US, Australia, Canada, and Japan, with a focus on e-commerce and payment infrastructure, according to a report published over the weekend by the security firm NorForge. The researchers discovered the activity while investigating publicly exposed AI-driven penetration-testing control panels.
NorForge did not name the victim organizations in its original report, but a report in Reuters, with which NorForge is coordinating disclosure, subsequently identified several. They included US-based farm machinery company John Deere, home appliance company SharkNinja, and engine manufacturer Cummins.
Extent of Cyber-Damage Remains Unclear
Specifically, NorForge identified three activity clusters across the exposed infrastructure, including two assessed to be malicious. Moreover, the activity appeared primarily to be financially motivated, with no obvious sign of state-sponsored intrusion activity, according to the post.
“The operation is massive, with over 1,100 attack projects spread over the 14 dashboards likely associated with this group,” Snorre Fagerland, a NorForge founding researcher, wrote in the post.
The activity included the exposure of “hundreds of potentially sensitive secrets and records showing how automated agents were instructed to identify vulnerabilities and establish persistent access to victim websites,” he wrote.
The reported consequences varied by victim, and the full extent of the damage remains unclear. NorForge confirmed that at least one attack had resulted in a Web shell being installed on a commercial website, potentially giving the attacker the ability to execute commands on the server, according to the post. NorForge also observed separate operations targeting credit-card data and cryptocurrency.
Of the victims reported by Reuters, John Deere said it had taken remedial action but had no indication that attackers accessed sensitive information, while SharkNinja said the incident had been contained and that it affected a standalone product-support website. Meanwhile, Cummins was investigating a potential attack on a third-party vendor’s website.
Cairn Security did not immediately return a request for comment by Dark Reading about what the company knows about the activity.
Using AI for Attack Coordination
Importantly, instead of just helping automate or speed up individual tasks, the attackers used AI agents to coordinate multiple stages of an attack as a connected workflow, according to NorForge. This allows attackers to run multiple projects in parallel and sustain operations over hours or days, reducing the need for constant human direction.
“This isn’t a single attack,” Tyler Oliver, CEO and founder of threat-hunting platform Huntbase, tells Dark Reading in an email interview. “It’s an exposure of attacker infrastructure, a window into the tools several groups with different motivations are using.”
What the activity confirms is something experts have known for some time, he says: that “attackers move much faster than defenders because they aren’t bound by the same processes. … They’ve adopted AI faster than their defensive counterparts, and that gap will only grow.”
The attackers used Cairn to task AI agents with finding exploitable weaknesses, including SQL injection, unsafe file uploads, insecure deserialization, authentication bypasses, and server-side template injection, according to NorForge. Some instructions went further, explicitly directing agents to deploy Web shells capable of providing ongoing access.
The operation also appeared to automate target selection, with projects including traffic-based target prioritization and browser automation to establish whether e-commerce websites could process orders.
AI Shifts Cyberattack Economics, Puts More Orgs in Sights
Businesses need to realize, and realize quickly, that “AI is changing the economics of cyberattacks more than the techniques themselves,” observes Rickard Carlsson, CEO of Detectify, in an email interview with Dark Reading.
“The attackers in this report were exploiting familiar weaknesses,” he says. “What’s different is that an AI agent can now pursue those weaknesses for hours with very little human involvement.” That doesn’t mean human intervention totally disappears, Carlsson observes, but using AI to automate tasks means attackers can run hundreds or thousands of attempts at relatively little cost, lowering the barrier to conducting attacks at scale.
They also can engage in this activity with little investment in their own infrastructure or tools, he adds. “NorForge estimates that an average task costs around five dollars, although the figure is uncertain,” Carlsson says. “When the cost of trying falls that dramatically, attackers can afford to target far more organizations, including those that might previously have been considered not worth the effort.”
Driving New Cyber Defense Strategies
Given the threats that AI-driven attacks present, organizations need to arm themselves with the technology to shore up defenses, but in a measured way at first, experts say. Strategies should include a security solution that uses AI to make deploying and reinforcing security basics — such as products that automate network fixes, vulnerability scanning, and patching — easier to apply, Huntbase’s Oliver says.
Detectify’s Carlsson suggests the same, advising organizations to “know every Internet-facing asset you own, test it continuously, and treat leaked credentials and keys as a priority, since those are exactly what these systems collected,” he says. “Then close the gap between finding and fixing.”
Only once an organization has a secure foundation, they can then adopt a more active and “even preemptive, defensive posture,” Oliver adds.
Investing in these kinds of defenses is not something to sleep on. As NorForge continues to work to notify victims and manage the fallout of the activity, the firm also is attempting to get the offending Cairn panels taken down, Fagerland said. However, he acknowledged that “this might not be possible” and “depends a lot on how well the hosters respond to takedown requests” — meaning organizations are far from being taken off target lists.
