Researchers easily trick Fortune-500 companies' AI agents into running arbitrary code — supply-chain attack via llms.txt guidance file illustrates how data has become code
· Source: tomshardware.com
In short
Researchers from Pandex demonstrated that AI agents from Fortune 500 companies can be tricked into running arbitrary malicious code through compromised llms.txt files — guidance files used by AI agents to understand how to interact with websites and software. The researchers found 237 references to outdated, mistyped, or abandoned software packages across 8,565 llms.txt files checked, and successfully got their own "malware" executed by agents within minutes of going live. The study highlights how the distinction between data and code is eroding with agentic LLMs, with frontier models like GPT-5 Luna and Sol running the malware over 90% of the time, illustrating a growing supply-chain attack vector in AI-driven software ecosystems.
Read Original Article — Read the full article on the publisher’s site.
In the issue of week 36, 2026: AI-Driven Automation and Integration Reshape Supply Chain Efficiency and Security
Tags: supply-chain, security, ai-agents, llms.txt, cybersecurity, vulnerability