Malware abuses AI coding-agent files to steal credentials and spread
Original title: Hackers Turn AI Coding Agent Configuration Files Into Malware Execution and Propagation Channels
Bitdefender found 22 malicious links whose setup.js loader can poison repositories and republish packages when stolen tokens allow it.
A campaign uses compromised GitHub content to target developers of AI coding assistants. Instruction files such as AGENTS.md and CLAUDE.md, or a session-start hook, can launch a concealed JavaScript loader. Bitdefender says the payload can steal source-control, cloud, CI/CD, registry, and SSH credentials, then alter local agent configurations. Planting malicious configurations elsewhere is a documented capability, not proof every target was compromised.
Unconfirmed: Documented propagation capabilities are not evidence that every targeted repository or registry was compromised.
- The loader searches configurations for 14 tools, including Claude, Codex, Copilot, and Cursor.
- It can tamper with hooks and Model Context Protocol server definitions.
- Noninteractive and SDK Claude Code sessions may not show the workspace trust prompt.
- Bitdefender says AI Guardian caught the instruction file in all 22 samples on November 10.
Impact: Stolen developer credentials and possible poisoned repositories or packages
- 1.Review agent instructions, startup hooks, MCP definitions, and editor tasks before trusting a workspace.
- 2.If execution is suspected, isolate the workstation and revoke tokens from a clean system.
- 3.Audit commits, workflow changes, and package releases made with the affected credentials.
- ATT&CK
- T1059.007 Command and Scripting Interpreter: JavaScriptT1552.001 Unsecured Credentials: Credentials In FilesT1195.002 Supply Chain Compromise: Compromise Software Supply Chain
- Vendors
- Anthropic
- Products
- Claude CodeCursorGitHub Copilot
- Malware
- setup.js
- Organizations
- Bitdefender
- AI models
- Claude
Indicators of compromise
auto-extracted · verify before use · export allAll| Type | Indicator | Context |
|---|---|---|
| md5 | 2191cbd04134839daeb6b8909725eae0 | vops ab7ff2f782c194953e5811d16b9e1817 Imanilom/baraja-coffe 2191cbd04134839daeb6b8909725eae0 Note: IP addresses and domains are intentionally defanged ( |
| md5 | 2f54f6d41f41bb3c8e684b0f009cbcc3 | 9a182b646c7fdff68f2ff9dfd822516 ConcordiaDev/theme_saimaniq 2f54f6d41f41bb3c8e684b0f009cbcc3 itsunai/docker-php-fpm 3794736979327349404171798328a38a eja |
| md5 | 3794736979327349404171798328a38a | niq 2f54f6d41f41bb3c8e684b0f009cbcc3 itsunai/docker-php-fpm 3794736979327349404171798328a38a ejaaduipitara/djp-devops ab7ff2f782c194953e5811d16b9e1817 I |
| md5 | ab7ff2f782c194953e5811d16b9e1817 | m 3794736979327349404171798328a38a ejaaduipitara/djp-devops ab7ff2f782c194953e5811d16b9e1817 Imanilom/baraja-coffe 2191cbd04134839daeb6b8909725eae0 Note |
| md5 | b9a182b646c7fdff68f2ff9dfd822516 | rs of Compromise Repository setup.js MD5 toray/QiniuSDK-lib b9a182b646c7fdff68f2ff9dfd822516 ConcordiaDev/theme_saimaniq 2f54f6d41f41bb3c8e684b0f009cbcc |
From the original article
3 min read at gbhackers.com
A malware campaign that uses compromised GitHub content to target developers using AI coding assistants. The malware can steal developer credentials, alter local agent configurations, and use stolen access to poison repositories that other developers may later open. The campaign exploits the trust coding assistants place in repository files such as AGENTS.
By Mayura Kathir · gbhackers.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.