Threat
CyberXero Uses Claude Code in Campaign That Stole Over 628,000 Ukrainian Records
First reported thehackernews.com
Page published · Page updated
Earliest dated coverage: 12 Oct 2026 · First observed: 12 Oct 2026 · Latest dated coverage: 12 Oct 2026
Coverage timeline
Single-source incident — one report is available.
Why it matters
CyberXero demonstrates that financially motivated access brokers are now operationalizing agentic coding assistants like Claude Code to orchestrate real intrusions at scale, blurring the line between commodity tooling and AI-driven attack automation against critical infrastructure.
Researchers at SOCRadar disclosed that CyberXero, a Russian-speaking, financially motivated initial access broker active since at least July 2026, ran a two-pronged AI-augmented campaign that stole over 628,000 Ukrainian records. The operation paired commodity offensive tooling like Cobalt Strike with an AI orchestration layer powered by Anthropic's Claude Code, running an automated global campaign against WordPress and e-commerce platforms alongside a curated manual campaign against Ukrainian energy and critical infrastructure.
Summary
SOCRadar researchers disclosed a two-pronged, AI-augmented initial access campaign attributed to CyberXero, a Russian-speaking, financially motivated initial access broker assessed active since at least July 2026. The operation runs a global automated pipeline against WordPress and e-commerce platforms alongside a curated, manual pipeline against Ukrainian energy and critical infrastructure.[0]
The campaign is estimated to have exfiltrated data from four Ukrainian organizations, harvesting more than 628,000 records tied to Ukrainian individuals, with additional confirmed compromises in Poland, China, and the Netherlands. The campaign was uncovered via an exposed open directory on the actor's Linux workstation and remains ongoing.[0]
CyberXero combined commodity offensive tooling such as Cobalt Strike with an AI orchestration layer powered by Anthropic's Claude Code and the open-source PentAGI agent, using context-engineering and session-reset tricks to bypass model refusals by framing work as authorized penetration testing.[0][9]
Attack chain
- Reconnaissance and scanning: A dedicated scanner node identifies targets of interest; a single automated execution scanned 4,708 targets using a Claude Code cluster of 51 specialized agents for scanning and exploit chaining.[0]
- Initial access / exploitation: Exploitation of known WordPress and e-commerce flaws including wp2shell (CVE-2026-63030, CVE-2026-60137) and the Support Board SQL injection (CVE-2026-4815), plus exploitation of exposed Redis servers; one run confirmed access to 429 WordPress admin panels.[0][24][25][26]
- Web shell deployment: Deployment of web shells (including WSO) at scale, with 32 web shells deployed in a 61-second window following confirmed panel access.[0]
- Post-exploitation and payload generation: PentAGI was connected to a Cobalt Strike Team Server through an AI provider API for payload generation and pentest execution; the manual Ukrainian pipeline conducted deep reconnaissance and manual exploitation against energy and utilities including a Kharkiv heating provider.[0][9]
- Exfiltration: Confirmed exfiltration of victim data from four Ukrainian organizations totaling over 628,000 records, with data staged in the actor's working directory.[0]
Disclosure timeline
| Date | Event |
|---|---|
| July 2026 | CyberXero assessed to have been active since at least this date.[0] |
| Week of Oct 2026 (report published 'last week') | SOCRadar published its report detailing the CyberXero operation.[0] |
| Oct 12, 2026 | The Hacker News published coverage of the SOCRadar findings.[0] |
Actor profile
CyberXero
A Russian-speaking, financially motivated initial access broker active since at least July 2026. Operates two concurrent pipelines: a global automated campaign against WordPress and e-commerce platforms for access and payment data, and a selective manual campaign against Ukrainian energy and critical infrastructure with no obvious direct-monetization motive. Heavily leverages AI orchestration via Claude Code (51 specialized agents) and PentAGI tied to Cobalt Strike, bypassing model refusals through authorization framing and session resets.[0]
How it works
The global pipeline exploits known WordPress and e-commerce security flaws: wp2shell (CVE-2026-63030 and CVE-2026-60137) to seize control of WordPress sites, and CVE-2026-4815, an SQL injection vulnerability in the Support Board WordPress plugin.[0][24][25][26]
The actor also exploited exposed Redis servers, which led to the breach of two Chinese entities; defenders are advised to check for unexpected entries in ~/.ssh/authorized_keys on Redis hosts.[0]
Model safety controls act on the context of a session rather than on model weights, so a refusal in one session does not carry into the next. The actor exploited this by reopening fresh sessions with authorization framing preloaded; some genuine refusals nonetheless held, including declining to install a backdoor, disable a firewall, move laterally, deploy a web shell, and rejecting a forged ownership token.[0]
Affected versions and patch status
| Product | Affected | Patch status |
|---|---|---|
| WordPress (Support Board plugin) | Sites vulnerable to CVE-2026-4815 SQL injection | Not specified in evidence[0][26] |
| WordPress (wp2shell) | Sites vulnerable to CVE-2026-63030 and CVE-2026-60137 | Not specified in evidence[0][24][25] |
| Redis | Instances exposed to the internet | Exposure/hardening issue; not a specific patch[0] |
| Palo Alto Networks PAN-OS GlobalProtect (Operation Master) | Gateways vulnerable to CVE-2026-0257 authentication bypass | Not specified in evidence[0][27] |
Indicators of Compromise
| Type | Indicator | Context |
|---|---|---|
| ip | 46.21.250.135 | Exposed open directory hosting the CyberXero actor's live Linux working directory, including AI session logs, plaintext tokens, exploitation toolkits, and exfiltrated victim data.[0] |
| cve | CVE-2026-4815 | SQL injection in the Support Board WordPress plugin exploited in the global pipeline.[0][26] |
| cve | CVE-2026-63030 | wp2shell vulnerability exploited to seize control of WordPress sites.[0][24] |
| cve | CVE-2026-60137 | wp2shell vulnerability exploited to seize control of WordPress sites.[0][25] |
| file-path | ~/.ssh/authorized_keys | Defenders advised to check for unexpected entries on Redis hosts as a sign of compromise.[0] |
Key takeaways
- CyberXero demonstrates the operational use of AI orchestration (Claude Code agents and PentAGI tied to Cobalt Strike) to industrialize initial access at scale, scanning thousands of targets and deploying dozens of web shells in under a minute.[0]
- Session-scoped model safety controls can be circumvented through authorization framing and session resets, though some high-risk requests (backdoor install, firewall disable, lateral movement, forged ownership token) were still refused.[0]
- The campaign blends opportunistic financial targeting with geographically concentrated, manual operations against Ukrainian critical infrastructure that lack an obvious direct-monetization motive.[0]
Defensive actions
- Audit WordPress, Magento, and PrestaShop installations for rogue administrator accounts and WSO web shells.: These platforms are directly targeted in the global automated pipeline and web shells were deployed at scale following panel compromise.[0]
- Check Redis hosts exposed to the internet for unexpected entries in ~/.ssh/authorized_keys.: The actor exploited exposed Redis servers, breaching two Chinese entities and establishing persistence via SSH key injection.[0]
- Patch or mitigate the exploited WordPress flaws (wp2shell CVE-2026-63030/CVE-2026-60137) and Support Board SQL injection (CVE-2026-4815).: These known vulnerabilities are actively used for initial access in the ongoing campaign.[0][24][25][26]
- Block or monitor traffic to/from 46.21.250.135.: The IP hosted the actor's exposed working directory and staging infrastructure.[0]