Top Tweets for #operationcairn

Operation CAIRN: An Autonomous AI Attack Platform Found Running in the Wild Potential Working name: UNC-CAIRN. Calling it: Operation CAIRN. Here’s what we found. It started as a routine IOC scan. A suspicious VPS on a Hong Kong-registered APAC hosting provider. Unusual port combination, a two-day self-signed TLS cert, a Python API behind Basic auth, and an open directory. That directory contained 150+ files from an active campaign. Inside was a custom-built platform called Cairn. Not a script. Not a toolkit. A full autonomous attack orchestration system powered by an LLM. The operator gives it a target and a goal. The AI does the rest Architecture: A dispatcher reads a SQLite target database. Each victim gets its own Docker container running Kali Linux. An LLM receives the target, a goal (“obtain webshell or RCE”), and a fact-graph of everything tried in prior sessions. It picks the next attack step, executes it, observes the result, and writes back into the graph. No human in the loop per session. The model is DeepSeek v4.1-flash, accessed through a pirated account rotation service called ClinePass that cycles stolen subscriptions to bypass per-session rate limits. Two confirmed active accounts. Combined remaining balance: roughly $10,000 USD. Scale across 13 days of logged activity: 229,828 AI-dispatched attack events 17,768 active exploitation attempts 85+ distinct targets in the database 2 Raspberry Pi devices used as local workers, named pi-workbuddy and pi-cline-deepseek What separates Cairn from automation is that the AI tracks failure states. Each session ends with a structured Chinese-language summary: what was tried, what was blocked, what is now confirmed. All of it written back into a fact-graph for the next session. Across 90+ sessions against a single target, the model is reading its own prior work. CVEs being weaponised: CVE-2024-34102 (CosmicSting, Magento XXE) CVE-2022-24086 (Magento template injection RCE) CVE-2026-75650 “StyleSmuggler” (Magento 0day at time of attacks) CVE-2026-53787 (Amasty Order Attributes unauth file upload) CVE-2014-6271 (ShellShock, the one that actually landed) CVE-2026-44277 (FortiAuthenticator pre-auth) ShellShock via SMTP is the most interesting vector here. The operator injected a bash payload into the RCPT TO field of a mail server. The target ran qmail, which strips the recipient domain after the @ before passing the local part to the delivery agent. That behaviour allowed shell execution through a normal SMTP handshake. Command output came back piped via curl, embedded inside 250 reply codes. The logs confirm it worked: EXEC [redacted]:25 2026-09-29T18:11:49 id→ 250 ok hostname→ 250 ok uname -a→ 250 ok Every command response smuggled back inside routine SMTP acknowledgements. The operator also compromised a multi-tenant SaaS platform running on GCP. A compromised Redis pod had this in its crontab, running every five minutes: */5 * * * * /bin/sh /data/pwn.sh The script was a netcat reverse shell calling back to the operator’s C2. It was still running at the time of discovery. Environment variables in the pod exposed internal service names for roughly 20 named customer tenants. Cairn is maintained software. It ships a memory management daemon called cairn_guard.sh. It monitors the dispatcher process, defers restarts until no attack sessions hold active leases, and rate-limits restart frequency to prevent loops. The comment in the script explains it was written after the dispatcher grew to 12 GB, filled swap, and was killed by the kernel’s OOM killer. Someone debugged that and fixed it. #ThreatIntel #CTI #OperationCAIRN