# AI-Agent-Driven Campaign (Strix/Hermes/Cairn) Compromises 27+ Retailers and Skims 600,000+ Payment Cards

> Gambit Security reports a financially motivated, apparently Chinese-speaking operator who used three open-source AI harnesses (Strix, Cairn, Hermes) to run mostly unattended intrusions against online retailers from July 2026. At least 27 companies were compromised, skimmers were placed on 119+ sites, and 600,000+ valid cards were taken from two companies.

- **Published:** 2026-11-14T10:49:12Z
- **Last reviewed:** 2026-11-14T10:49:12Z
- **Canonical:** https://intel.threadlinqs.com/threat/TL-2026-6069
- **Card image:** https://intel.threadlinqs.com/og/threat/TL-2026-6069.png
- **ID:** TL-2026-6069
- **Severity:** HIGH
- **Category:** THREAT_INTEL
- **Status:** ACTIVE
- **Detections:** 9 (1 shown free below) · **IOCs:** 28 (3 shown free below)

## Description

Gambit Security recovered the operator's exposed staging server and published an interim report on 22 September 2026 (covered by BleepingComputer, The Register and the Cloud Security Alliance). The campaign began no later than July 2026 and was still active when reported. No named threat group is attributed. The operator wrote 1,951 short Chinese-language prompts across 260 Hermes sessions, and model access ran through OpenRouter.

The tooling has three roles. Strix, an open-source AI penetration-testing tool, ran 146 times in deep mode against 138 hosts between 23 and 31 August 2026 (633 scanner hours over 195 clock hours), first on GLM 5.2 and then on DeepSeek v4 Pro. Cairn, an autonomous exploitation engine given an objective such as a shell or admin access, ran on DeepSeek v4.1 Flash. Hermes, an open-source autonomous agent with persistent memory, self-written skills, scheduled jobs and a web console, orchestrated the campaign. It loaded a 'SOUL - Red Team Operator' persona and 121 skills (78 attack-related), including one that removes the harness's content filters. Hermes ran on Anthropic's claude-opus-4.6; Gambit says newer models refused the requests. The Cairn in this campaign is not the Cisco Talos malware-analysis tool of the same name.

Targets came from a website traffic-ranking service, filtered to shops running custom code rather than major commerce platforms. Some targets were chosen by hand, and some were handed to the agent with a working admin password. Access usually took under a day, often a few hours. Between 10 and 15 September 2026, 105 attack projects were launched (48 visible, 57 deleted) and at least 27 companies were compromised to varying degrees. One documented Cairn chain: unauthenticated error-based SQL injection (EXTRACTVALUE) in a login email parameter, then a plaintext OTP read to bypass MFA, then the admin panel, an unrestricted file upload to host RCE, sudo NOPASSWD python3.12 to root, an NFS mount with no_root_squash, WordPress DB credentials from wp-config.php, a WordPress plugin upload for RCE, an AWS Secrets Manager dump (46 secrets), Magento Aurora DB access, Magento encryption key extraction, and Blowfish-ECB decryption of stored card numbers. Each chain was chosen in real time, so TTPs varied by victim.

Skimmer deployment depended on access: a loader appended to legitimate JavaScript files with timestamps restored; foreign script tags on checkout pages pointing to look-alike domains; inline code inside Google tag (gtag) blocks padded with tabs; S3/CDN bucket poisoning through an exposed AWS key; loader stored in database product-description fields and later moved to a file; a Kubernetes initContainer added to the production front-end deployment; and server-side page-cache poisoning of a checkout page model. A re-infection cron job in the JBoss log directory checked file size every two minutes and re-appended the skimmer if it was reverted. The loader is a base64 payload with a seven-character junk prefix, run as new Function(atob('...').slice(7))(), which creates a script element pointing to a remote file such as nrt.js.

Impact: more than 600,000 unexpired cards from two companies (about 79%, 488,372, US-issued) were validated by Overwatch Data, which handled issuer notification. A payment processor found at least 60% of a sampled batch had not previously been flagged for fraud. Skimmers were ordered against at least 27 named victims and confirmed on 19. Researcher Varys helped find 100+ more infected sites, which gives the 119+ total. Some level of access was reported at a Fortune 500 hospitality company, a major US airline, a large industrial supplies distributor and a US online fashion retailer. A Hermes 'Database Wipe After Extraction' skill, intended to clear source card fields from Magento databases, caused data loss at several retailers. At one bicycle retailer an over-broad name match dropped 180 tables, including backups. Gambit puts the OpenRouter spend at $7,005.71 over the first four weeks (captured 25 August) and total cost at $12,000-$18,000. The operator's own review gives a mean of $25.46 per completed scan across 101 scans (range $3.13-$79.31). Proxy providers used were IPRoyal, 711proxy and 1024proxy. Gambit calls the report interim and warns the tallies may be inaccurate. One search summary says Gambit worked with the Shadowserver Foundation on infrastructure takedown, but The Register and CSA do not confirm it.

## MITRE ATT&CK

- T1595.002 Vulnerability Scanning
- T1583.001 Domains
- T1588.007 Artificial Intelligence
- T1190 Exploit Public-Facing Application
- T1078.004 Cloud Accounts
- T1059.007 JavaScript
- T1505.003 Web Shell
- T1053.003 Cron
- T1548.003 Sudo and Sudo Caching
- T1027 Obfuscated Files or Information
- T1070.006 Timestomp
- T1552.001 Credentials In Files
- T1555.006 Cloud Secrets Management Stores
- T1056.003 Web Portal Capture
- T1005 Data from Local System
- T1090 Proxy
- T1048 Exfiltration Over Alternative Protocol
- T1485 Data Destruction
- T1565.001 Stored Data Manipulation

## Free detection (Splunk SPL, Microsoft KQL, Sigma)

**Connections to Campaign C2, DNS-Exfil and Skimmer Infrastructure - C2/Exfiltration** (`TL-2026-6069-SPL-003`)

TL-2026-6069: matches DNS, proxy or firewall traffic to published campaign infrastructure: medbooksource.com (console and DNS exfiltration), traffic-analyzer.net and 213.21.239.62/172.245.224.188 (C2), and look-alike skimmer hosts such as cdn.netlfjs.com, static-js.com and b8t.shop that serve skimmer scripts to checkout pages.

- **Languages:** Splunk SPL, Microsoft KQL, Sigma
- **ATT&CK:** T1090, T1048, T1583.001, T1056.003, T1059.007
- **Data sources:** DNS logs (index=network sourcetype=stream:dns), firewall/proxy (pan:traffic, zscaler:nss:web), DeviceNetworkEvents, DnsEvents
- **Severity:** high
- **Confidence:** high

### Splunk SPL

```spl
index=network (sourcetype="stream:dns" OR sourcetype="pan:traffic" OR sourcetype="zscaler:nss:web")
| eval host_ioc=lower(coalesce(query, url, dest_host, ""))
| eval dst=coalesce(dest_ip, dest, "")
| where match(host_ioc,"(^|[./])(medbooksource[.]com|traffic-analyzer[.]net|b8t[.]shop|netlfjs[.]com|x1opay[.]co|static-js[.]com|js-static[.]com|jsnetlify[.]com|netlifyjs[.]com|newssjs[.]com)") OR match(dst,"^(155[.]254[.]22[.]215|209[.]126[.]4[.]170|213[.]21[.]239[.]62|172[.]245[.]224[.]188|172[.]245[.]89[.]137)$")
| eval ioc_hit=if(match(host_ioc,"medbooksource[.]com"),"dns-exfil/console",if(match(host_ioc,"traffic-analyzer[.]net"),"c2","skimmer-or-c2"))
| stats count dc(host_ioc) as distinct_names min(_time) as first_seen max(_time) as last_seen values(ioc_hit) as ioc_hit values(host_ioc) as names by src, dst
| convert ctime(first_seen) ctime(last_seen)
```

### Microsoft KQL

```kql
let iocDomains = dynamic(["medbooksource.com", "traffic-analyzer.net", "b8t.shop", "netlfjs.com", "x1opay.co", "static-js.com", "js-static.com", "jsnetlify.com", "netlifyjs.com", "newssjs.com"]);
let iocIPs = dynamic(["155.254.22.215", "209.126.4.170", "213.21.239.62", "172.245.224.188", "172.245.89.137"]);
union
(DeviceNetworkEvents
 | where Timestamp > ago(24h)
 | where RemoteIP in (iocIPs) or RemoteUrl has_any (iocDomains)
 | project TimeGenerated=Timestamp, Source="DeviceNetworkEvents", Device=DeviceName, Process=InitiatingProcessFileName, Indicator=coalesce(RemoteUrl, RemoteIP)),
(DnsEvents
 | where TimeGenerated > ago(24h)
 | where Name has_any (iocDomains)
 | project TimeGenerated, Source="DnsEvents", Device=Computer, Process="", Indicator=Name)
| order by TimeGenerated desc
```

### Sigma

```yaml
title: DNS Query to AI Agent Retail Campaign Skimmer and C2 Domains - TL-2026-6069
id: e5a92d14-8b73-4c0e-a6f9-3d7b1c4e8901
status: experimental
description: Detects DNS resolution of domains attributed to the Strix/Cairn/Hermes campaign, namely the operator console and DNS-exfiltration domain medbooksource.com, C2 traffic-analyzer.net, and look-alike skimmer hosts such as netlfjs.com, static-js.com and js-static.com.
references:
    - https://gambit.security/blog-posts/autonomous-ai-agents-online-retailers-25-a-company
    - https://www.bleepingcomputer.com/news/security/malicious-ai-agents-steal-600k-credit-cards-infect-100-plus-sites-with-skimmers/
author: Threadlinqs Team
date: 2026-11-14
tags:
    - attack.command-and-control
    - attack.t1090
    - attack.exfiltration
    - attack.t1048
    - attack.resource-development
    - attack.t1583.001
    - attack.collection
    - attack.t1056.003
    - attack.execution
    - attack.t1059.007
logsource:
    category: dns
detection:
    selection:
        query|endswith:
            - 'medbooksource.com'
            - 'traffic-analyzer.net'
            - 'b8t.shop'
            - 'netlfjs.com'
            - 'x1opay.co'
            - 'static-js.com'
            - 'js-static.com'
            - 'jsnetlify.com'
            - 'netlifyjs.com'
            - 'newssjs.com'
    condition: selection
falsepositives:
    - Retrospective lookups by threat researchers
    - Sinkholed or repurposed domains after takedown
level: high

```

**Known false positives:**

- Retrospective lookups by threat researchers
- Sinkholed or repurposed domains/IPs after takedown

**Tuning notes:** Refresh the IOC lists from the TL-2026-6069 feed, enrich with browser-side CSP reports for checkout pages, and alert at higher severity if the source is a payment or storefront server. No community-corroborated or sibling indicators were available from OSINT context.

### 8 more detection rules (Splunk SPL, Microsoft KQL, Sigma), unlock with TL Creds

- Error-Based SQLi (EXTRACTVALUE/UPDATEXML) Against Retail Login - Exploitation (KQL) (locked: unlock with TL Creds)
- Sudo python3.12 Root Escalation and JBoss Log-Dir Cron Re-Infection - Post-Exploitation (KQL) (locked: unlock with TL Creds)
- Connections to Campaign C2, DNS-Exfil and Skimmer Infrastructure - C2/Exfiltration (KQL) (locked: unlock with TL Creds)
- Error-Based SQLi (EXTRACTVALUE/UPDATEXML) Against Retail Login - Exploitation (Sigma) (locked: unlock with TL Creds)
- Sudo python3.12 Root Escalation and JBoss Log-Dir Cron Re-Infection - Post-Exploitation (Sigma) (locked: unlock with TL Creds)
- Connections to Campaign C2, DNS-Exfil and Skimmer Infrastructure - C2/Exfiltration (Sigma) (locked: unlock with TL Creds)
- Error-Based SQLi (EXTRACTVALUE/UPDATEXML) Against Retail Login - Exploitation (locked: unlock with TL Creds)
- Sudo python3.12 Root Escalation and JBoss Log-Dir Cron Re-Infection - Post-Exploitation (locked: unlock with TL Creds)

Sign up free to unlock the detection rules: https://threadlinqs.com/?modal=signup

## Sample indicators (3 of 28)

Defanged, TLP:CLEAR.

- IP address: `155.254.22[.]215`
- Domain: `medbooksource[.]com`
- IP address: `209.126.4[.]170`

Full set of 28 indicators unlocks with TL Creds: https://threadlinqs.com/tl-creds

## Sources

- [Malicious AI agents steal 600K credit cards, infect 100+ sites with skimmers (BleepingComputer)](https://www.bleepingcomputer.com/news/security/malicious-ai-agents-steal-600k-credit-cards-infect-100-plus-sites-with-skimmers/)
- [Gambit: Autonomous AI agents, online retailers, $25 a company](https://gambit.security/blog-posts/autonomous-ai-agents-online-retailers-25-a-company)
- [The $25 Breach: Autonomous AI Agents Run Skimming Rings (Cloud Security Alliance research note)](https://labs.cloudsecurityalliance.org/research/csa-research-note-ai-agent-skimming-campaign-20260926-csa-st/)
- [AI agents used in credit card theft campaign (The Register)](https://www.theregister.com/a/5299012)
- [AI-powered campaign targets hundreds of online retailers (SecurityWeek)](https://securityweek.com/ai-powered-campaign-targets-hundreds-of-online-retailers)
- [AI agents stole 600,000 credit cards for about $25 a target, Gambit says (The Next Web)](https://thenextweb.com/news/ai-agents-600000-credit-card-records-gambit)
- [Massive Chinese hack uses AI agents to steal over 600,000 credit cards (TechRadar)](https://techradar.com/pro/security/massive-chinese-hack-uses-ai-agents-to-steal-over-600-000-credit-cards-and-hit-hundreds-of-sites-with-malware)

## Full data

The other 8 detection rules (Splunk SPL / Microsoft KQL / Sigma) unlock with TL Creds; a free account includes daily creds: https://threadlinqs.com/?modal=signup
Indicator values unlock with TL Creds, at a lower price on Red and Purple; AI agents can query rules and IOCs through the Threadlinqs MCP server (Purple): https://intel.threadlinqs.com/mcp

Canonical: https://intel.threadlinqs.com/threat/TL-2026-6069
