1. The URL guard only classified IP-literal hostnames; any plain domain
name (the common case) skipped validation entirely. Hostnames are now
resolved via DNS and every returned address is checked against the
same private/loopback/link-local ranges (net.BlockList), so an
attacker-controlled domain pointed at an internal IP is rejected too.
2. fetch() follows redirects by default, so a registered public URL could
302 to an internal target at delivery time. Delivery now uses
redirect: 'manual' so such a redirect is never followed.