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
TL;DR
Security researchers got AI agents at Fortune 500 companies to execute arbitrary code. The attack ran through llms. txt, the public guidance file many sites publish for AI systems. Instructions planted in that file were executed as commands rather than read as data.
Nauti's Take
The clean disclosure is the opportunity here: llms. txt can be hardened before attackers scale the pattern.
The risk stays high, because many agents still treat fetched web content as instructions. Teams running agents with tool access should handle external context files like untrusted code and cut write and execution permissions to the minimum.
Sources
