Cases › security breach › LR-2026-0902
Malware campaign abusing AI coding assistant instruction files and session-start hooks to run setup.js
- How it was found
- Outside researcher
- Reported
- 17 Nov 2026
- Severity
- Moderate
- An AI coding assistant reads project instruction files such as AGENTS.md and CLAUDE.md to understand a repository and carry out developer tasks. Such an agent can have access to tools, shell commands, credentials, local files, or cloud services.
- The planted instructions told the assistant to run scripts/setup.js, a file it was told not to read, and to run it without asking the user for confirmation. This crossed the boundary of not executing unreviewed code on the developer's machine.
- The malware can steal credentials, validate which accounts it can access, modify repositories, infect software packages, and plant new AI-agent instructions. Bitdefender Labs had identified 22 malicious GitHub links so far.
- Bitdefender Labs researcher Andrei Anton-Aanei uncovered the campaign and identified 22 malicious GitHub links.
Source: Bitdefender
What was done
Monitoring added (Developer or vendor)
Risks Hidden instructions hijack the agent (66 incidents) · AI used to run a cyber attack (97 incidents) · AI system attacked or breached (54 incidents)
No source states: AI company and model, money lost or records, tools it had, company using the AI, sector.
Evidence
- Cause hostile input carried instructions
Sources (1)
- AI Agents Targeted Through Malicious GitHub Repositories · Bitdefender · 2026-11-17
Show every field, with its source and grade
What happened
Earliest source we hold
2026-11-17
D inferred by us from the described behaviour, reason shown
Who caught it
an independent researcher or journalist
D inferred by us from the described behaviour, reason shown
Cause
hostile input carried instructions
D inferred by us from the described behaviour, reason shown
Cause in detail
Untrusted repository instruction files carried the instruction to run setup.js.
D inferred by us from the described behaviour, reason shown
Where its input came from
Malicious GitHub repository content, including instruction files such as AGENTS.md and CLAUDE.md
C one secondary source states it
What went wrong
security breach
D inferred by us from the described behaviour, reason shown
What it did
Malicious instructions in repository instruction files persuaded the assistant to execute setup.js.
D inferred by us from the described behaviour, reason shown
Impact
Severity
moderate
D inferred by us from the described behaviour, reason shown
Kind of harm
others' security
D inferred by us from the described behaviour, reason shown
Kinds of data exposed
credentials
D inferred by us from the described behaviour, reason shown
Why this severity
Other parties' data or systems were touched, and a source states how far it reached. The malware can steal credentials and modify repositories and packages of others, with 22 malicious GitHub links identified, a bounded scope.
D inferred by us from the described behaviour, reason shown
Notes on gaps
The title's session-start hooks are not covered by the answers. The done step is not answered by the source.
LR-2026-0902 · version 2 · published 2026-11-17 after automatic checks; not yet checked field by field by a person