AI · 2h ago
BleepingComputer reports researchers demonstrated how a malicious AGENTS.md file in one customer's Downloads folder steered an AI coding agent into installing a rogue npm package.
AGENTS.md files provide instructions to coding agents. Here, the agent followed instructions outside the project it was working on, then installed the package. The repository did not need to contain the hostile instruction for the agent to introduce software into the developer's environment.
For teams whose assistants can read local files beyond a checked-out project, Downloads is part of the agent's instruction surface too. A clean repository review may not reveal why an agent added a package, and that package can carry risk into local development or build workflows.
4 sources covering this story
AI coding agents fall for malicious instructions planted in the Downloads folder - Help Net Security
A planted AGENTS.md file tricked OpenAI’s Codex into installing malware, exposing a risk for AI coding agents that act on local files.
Malicious downloads can trick AI coding agents into compromising themselves
Security researchers found that trusted instruction files can be abused to install malware, redirect API traffic, and weaken built-in safeguards in popular coding assistants.
Malicious File in Downloads Folder Tries to Trick AI Coding Agent Into Installing Malware
A malicious AGENTS.md file led Codex to execute npm malware in over 90% of tests, while a separate Claude Code attack exposed login tokens.
AI coding agents hijacked using AGENTS.md file in Downloads folder
Researchers found a malicious AGENTS markdown instruction file in a customer's Downloads folder that tricked an AI coding agent into installing a rogue npm package.
Part of the PlainSec briefing for 2026-10-26