The Pulse
AI Agents Helped a Hacker Breach More Than 100 Companies
NorForge found more than 20 exposed Cairn panels tied to likely malicious activity and documented an operation targeting e-commerce sites in four countries. Reuters identified more than 100 companies whose websites were breached, including
AI.info Team · Written with AI · fact-checked ·
AI agents attempted to break into more than 1,200 websites and succeeded against sites belonging to more than 100 companies in a campaign that cost an estimated $5,000 in model tokens, Reuters reported on October 19. The targets included companies in the United States, Canada, Australia and Japan. NorForge’s investigation of exposed Cairn control panels gives a separate view of the operation—and of the security lapses that let outsiders watch it.
More than 1,100 projects in one cluster
NorForge said it found more than 20 publicly exposed Cairn panels likely linked to malicious activity. Its largest cluster, which it called AIA0001, used 14 panels and contained more than 1,100 attack projects. The firm assessed the cluster as financially motivated, with targets concentrated in e-commerce and payment infrastructure.
Cairn is a state-based problem-solving framework: an operator sets a starting point and a goal, while agents record discoveries and decide what to try next. NorForge found that the operators’ target-selection pipeline used Playwright, a browser automation tool, to check that shopping sites could accept orders before agents received them. A typical task sought a route to unauthorized server access, according to the firm’s translation of the instructions.
Reuters independently reviewed Cairn activity and identified Deere, SharkNinja and Cummins among the companies whose websites were breached. Deere told Reuters it had taken “necessary remediation actions” and had no indication sensitive information was accessed. SharkNinja said the affected site was a standalone product-support service and the threat had been contained; Cummins said it was investigating a possible attack on a third-party vendor’s website.
Agents worked for hours without close supervision
NorForge estimated an average success rate of about 10% across the campaigns it examined, while cautioning that the different operators and definitions of success make comparisons difficult. For successful AIA0001 projects, the average time from start to goal was just under 23 hours, and each generated about 140 recorded facts. Reuters reported that the Chinese-speaking operator’s campaign attempted more than 1,200 site intrusions and that Snorre Fagerland, NorForge’s principal threat researcher, estimated roughly $50 in token costs per successful hack.
“It’s a new world now.”
Snorre Fagerland, principal threat researcher at NorForge, speaking to Reuters
The exposed panels raised a second risk beyond the intrusions themselves. NorForge found hundreds of secrets in the panel data, with 11% verified as valid by its Trufflehog scan. The interfaces also appeared to expose controls for starting or stopping projects and redirecting agents, but NorForge said it did not test whether those controls worked without authorization.
Exposed dashboards put clients at risk
NorForge said the panels revealed target identities, vulnerabilities, credentials and whether attacks had succeeded. The company redacted sensitive details and withheld panel addresses while seeking to have them taken down. It also warned that apparently legitimate red-team operations had left customer tests exposed through unsecured dashboards.
The report describes successful intrusions, but it does not establish how much information attackers took. Reuters likewise said it could not determine the extent of unauthorized data extraction. For Deere, the company said it had no indication sensitive information was accessed; for the wider set of victims, the amount of data taken remains unresolved.