AI Agents Reached 100+ Companies: FinOps Lessons
Updated October 21, 2026 · first published October 21, 2026
A campaign reported by NorForge and Reuters used AI agents coordinated through exposed Cairn panels to probe more than 1,200 websites and reach sites associated with more than 100 companies. Those are different evidence sets: public reporting does not establish that every target was fully compromised or that data was taken from each. The FinOps lesson is practical: inexpensive automated attempts can create much larger response and recovery costs when basic web controls fail.
Separate attempts from confirmed impact
NorForge researcher Snorre Fagerland found exposed Cairn instances being used maliciously and could observe project records. Reuters reviewed attacker activity and independently identified some targets, including Deere, SharkNinja, and Cummins. Companies described limited or still-investigated impact; NorForge also confirmed individual compromises. The reports distinguish 1,200+ sites attempted from 100+ companies reached. Do not present either count as 1,200 confirmed breaches.
Automation changes attacker economics, not security fundamentals
Cairn coordinates agents toward a goal and records findings on a shared board; it was not built solely for abuse. Reported goals included code execution and persistent access on commercial websites. NorForge estimated roughly 10% success across observed attempts, depending on which clusters and goals are counted. Reuters cited the researcher’s rough estimate of about $5,000 in campaign token spend, or around $50 per successful intrusion. These are attributed estimates, not audited invoices.
The exposed control panel was a second leak
Some Cairn panels were internet-accessible and exposed targets, notes, and secrets. NorForge reported hundreds of secrets in exposed data and verified 11% were still valid. An agent orchestration system can leak credentials even when a target was not compromised. Keep dashboards, task boards, tool logs, and outputs private; minimize retained secrets, rotate credentials found in logs, and monitor unexpected access.
Budget for defense and response, not just tokens
Patch public-facing frameworks and plugins, remove unused endpoints, protect admin panels, and apply least privilege to application and database identities. Monitor unexpected files, web shells, session-store writes, admin sessions, and outbound connections from web workers. Estimate triage, forensics, credential rotation, storefront downtime, fraud review, customer communication, and rebuild costs; reserve capacity for concurrent cases. An agent budget can bound authorized testing but cannot fix a vulnerable public app. Pair the agent budget framework with the spend-spike runbook.
FAQ
Did Cairn itself have a vulnerability?
The reporting describes exposed Cairn panels being abused and Cairn being used to coordinate malicious work. It does not establish that a Cairn product vulnerability caused the victim-site compromises.
Were all 1,200+ target sites breached?
No. The reports distinguish attempted sites from companies reached. Impact and data theft were not confirmed for every target.
How much did the campaign cost the operator?
NorForge and Reuters cited rough token-spend estimates, including about $5,000 for the observed campaign and roughly $50 per successful intrusion. These are not verified billing records.
Sources
- NorForge: Inside Cairn and exposed AI-driven cyber operations
- Reuters: How an attacker used AI agents to break into websites
Related
Related
Want this applied to your stack? Bring the provider bills, gateway logs, and top workflows; we will map the cost drivers and savings path. Book a free audit →