Security · Updated 25 Oct, 09:19 pm IST
AI coding agents hijacked via malicious AGENTS.md files
Why it matters for readers: It matters because plain-text project instructions can be abused to make AI tools do harmful things without executable malware.
- Forever Security discovered a malicious AGENTS.md in a customer’s ~/Downloads that caused an agent to install a rogue npm package named 'function-color'.1
- Agents automatically load applicable files from their working directory and can follow instructions in plain-text files like AGENTS.md to run commands or install packages.1
- Researchers reproduced the attack against OpenAI Codex and showed a related technique against Anthropic’s Claude Code to exfiltrate OAuth tokens and hijack accounts.1
- AGENTS.md files are not executables but can contain project-specific instructions that, if malicious, cause agents to misuse available tools and permissions.1
Get a brief like this every morning
Uzha reads hundreds of sources and gives you the stories that matter for your work, with every source linked. Free.
Get started