AI coding agents followed abandoned package references, 6K domains analyzed

Researchers analyzed over 6,000 domains and found 120 unclaimed package or domain references in llms.txt files used by AI coding agents. These stale references could allow attackers to register the names and deploy malicious packages, potentially compromising corporate systems. An experiment showed that Fortune 500 companies' systems contacted these malicious packages within an hour, highlighting a new software supply-chain attack vector.