Skip to content
NewA calmer, faster ZeroHour
StoryMalwarehigh

Malicious GitHub files steer AI coding agents to steal secrets

3 sources · 3 articlesFirst reported Updated ()1
What's new

November 18: GBHackers adds that the loader searches 14 tools, can alter MCP definitions, and that noninteractive Claude Code sessions may skip the trust prompt.

Merged summary · x-ai/grok-4.7 · rewritten as coverage arrivesSummaries by Vani AI

Bitdefender found 22 malicious GitHub links that can make AI coding agents run a loader, steal secrets, and spread.

Bitdefender Labs says a GitHub campaign targets developers using AI coding assistants. Files such as AGENTS.md and CLAUDE.md, or session-start hooks such as .claude/settings.json, can launch scripts/setup.js after a repository is trusted. The loader is a 3.75 MB multi-stage JavaScript file that decodes, decrypts, and runs a bundle through Bun, then steals CI/CD, cloud, registry, browser, SSH, and password-manager secrets. It can rewrite agent settings and, using stolen GitHub or GitLab credentials, push the trap into more repositories or try to republish packages. Bitdefender says AI Guardian detected all 22 samples in internal tests on November 10.

Key facts
  • Researcher Andrei Anton-Aanei identified 22 malicious GitHub links.
  • Outer script uses hex, ROT13, and AES-128-GCM; extra payloads sit in package.json.
  • Spread is capped at 25 repositories and 50 branches by default.
  • Loader searches configs for 14 tools and can tamper with hooks and MCP definitions.
  • IoCs include repositories under toray, ConcordiaDev, itsunai, and ejaaduipitara.
Analyze this story with Vani AIImpact, who is affected and what defenders should do, in a new chat

Timeline

dated events in the reporting · oldest first
  1. Nov 10, 2026AI Guardian detected all 22 samples in internal tests. · Bitdefender Labs

What each outlet adds

beyond the shared facts
  • Bitdefender LabsNames the researcher and the 3.75 MB loader techniques.
  • Cyber Security NewsGives spread caps, IoC repository names, and rotation advice.
  • GBHackersNotes 14 tools, MCP tampering, and skipped trust prompts.

Coverage

oldest first · one row per article · UTC
  1. · 1d ago Bitdefender LabsFirst report

    GitHub campaign steers AI coding agents into credential-stealing malware

    Bitdefender found 22 malicious links whose setup.js chain steals tokens and plants new agent instructions to spread.

  2. · 11h ago Cyber Security News

    Poisoned GitHub repos hijack AI coding agents to steal credentials

    Bitdefender found 22 malicious links; a loader steals tokens and can push the same trap into other repositories.

  3. · 8h ago GBHackers

    Malware abuses AI coding-agent files to steal credentials and spread

    Bitdefender found 22 malicious links whose setup.js loader can poison repositories and republish packages when stolen tokens allow it.