AI coding agents hijacked using malicious instruction files
Researchers discovered a malicious markdown file in a Downloads folder that tricked an AI coding agent into installing a rogue npm package.
Cybersecurity researchers have uncovered a novel security risk affecting AI-powered coding agents, demonstrating how local instruction files can be manipulated to compromise development environments.
According to a report by BleepingComputer, investigators found a malicious AGENTS markdown instruction file residing in a customer's Downloads folder. The rogue file successfully tricked the AI coding agent into downloading and installing an unauthorized npm package.
This type of attack highlights the inherent vulnerabilities of autonomous AI systems that rely on interpreting local files to execute tasks. Because these agents are designed for automation and convenience, they can be easily manipulated by hidden instructions if proper validation is missing.
As developers globally and in emerging tech hubs adopt AI coding assistants, such security incidents emphasize the critical need for strict access controls and heightened awareness regarding local file environments.
Experts advise developers to carefully audit configuration and instruction files, particularly those originating from untrusted sources or stored in general directories, before allowing AI agents to process them.
