In the wild
Realized harm
Exposed Cairn consoles reveal a hacker's mostly autonomous agents breaking into 100+ company websites
AIR-2026-0282 AIR record · 2026-10-norforge-cairn-agent-website-breaches
A real party was affected. This incident caused actual consequences, not a demonstrated capability.
From the supporting source
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.theverge.com
The record
Disclosed
2026-10-19
Vendor
multiple victim organisations
Product
Cairn open-source agent platform (misused by the operator)
Risk domain
Access control and permissions › Privilege escalation
System type
Generative model
Who acted
Agent acted (contested) — the operator set a goal and the agents carried out the intrusions largely on their own, visible in their own progress logs; a human defined start and end points
Attack surface
Agent framework
Autonomy level
Semi-autonomous
Initial vector
Operator tasked autonomous intrusion
Impact type
Unauthorized access
Confidence
Medium
Submitted by
EAI Internal
Crosswalk
- OWASP Agentic Top 10 (2026)
- ASI03 – Identity and Privilege Abuse
- MITRE ATLAS
- AML.T0055 – Unsecured Credentials
- NIST AI RMF
- GOVERN 6.1
- EU AI Act
- Article 13
- ISO/IEC
- TR 24028; 42001; 23894
- AIUC-1
- B. Security
- Scope
- tool misuse
- Mitigation
- Apply least-privilege principle; Regular policy audits.
- Taxonomy
- Enkrypt AI agent risk taxonomy · Black-Box Red Teaming of Agentic AI: A Taxonomy-Driven Framework for Automated Risk Discovery
Blast radius
More than 100 websites across the US, Canada, Australia and Japan broken into by the most active operator; Norforge counted more than 20 exposed Cairn instances judged likely malicious
Capability abused
Defense & remediation
Guardrail kinds
Unknown
Guardrail outcome
Silent
Fix shipped
Norforge working with hosts to take the malicious instances down; SharkNinja says the affected site is contained
Taxonomy & references
Curator notes
Norforge found the campaign because the operators left their Cairn consoles reachable on the open internet, where the agents record their own step-by-step progress — a direct view of agents carrying out intrusions rather than an inference drawn from attack speed. The most active operator broke into more than 100 sites and wrote its preferred tactics in fewer than 100 Chinese characters; a successful break-in took about a day. The victim names come from Reuters as relayed by The Verge, not from Norforge: SharkNinja confirmed a breach of a standalone product-support site and says it is contained, John Deere reports no sign that sensitive data was reached, and Cummins is looking into a vendor-run site. The detail here is quoted from The Verge's account. A separate headline the next day put the figure above 1,200 targeted sites, which this reporting does not support. Cairn's author maintains the engine was never meant for hacking.
Supporting sources
Cite as
Enkrypt AI. "Exposed Cairn consoles reveal a hacker's mostly autonomous agents breaking into 100+ company websites". AIR, AIR-2026-0282. https://agentincident.vercel.app/i/AIR-2026-0282/ (accessed 2026-11-10).