A crawler will typically preprocess to remove the HTML comments before processing the document, specifically for reasons like this (avoiding prompt injection). So an LLM generating the summary would probably never have seen the comments at all.

So it's likely an actual person actually was looking at the full content of the document and the summary manually.