Check this site (the button in the AI panel, or ☰ → Check this site's safety…) answers questions like "is this a scam?" or "will I get malware from this download?" — but grounded in checks Vodou itself already ran, not the model guessing from general knowledge.

What it's told

Starting a check hands the model a plain summary of what Vodou already knows about the current page: whether the connection is HTTPS, whether the certificate is trusted, whether the address matches a known look-alike/typosquat pattern, whether it's on the local malicious-site list, and whether any of this session's downloads from that site look like an executable/installer. The model is explicitly told it has no antivirus scan, no live reputation lookup, and no way to browse further — so it answers from those facts instead of pretending to more certainty than a local check can offer.

A separate, one-time action

This is the one deliberate exception to ordinary Ask chat never attaching page content (see Summarizing search results and asking questions). It only runs when you click it, starts a fresh conversation rather than joining your existing chat, and you can keep asking follow-up questions afterward with those facts already in context.