Websites Are Embedding Hidden Instructions for AI Agents, and One Publisher's Bots Flagged Them
A developer who uses AI agents to publish articles across platforms reported that the agents encountered text on Hashnode, Indie Hackers, and Product Hunt pages that was addressed to AI bots rather than human readers. The agents were configured to treat webpage text as data only, so they ignored the embedded content, completed their assigned task, and reported what they found. The developer linked this to prompt injection, a risk named by Simon Willison in 2022 and ranked first on the OWASP Gen AI Security Project's 2025 list. OWASP notes such content can be invisible to humans yet still be parsed by models, and that no perfect defense may exist. Upon review, the developer found the publicly visible AI-directed files on these sites, such as llms.txt guides, were largely benign, though the situation highlights the need to keep outside content separate from genuine instructions.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.


Discussion (0)
Log in to join the discussion and vote.
Log in