Read this part more carefully:

By pushing routes that are more specific than a /0 CIDR range that most VPNs use, we can make routing rules that have a higher priority than the routes for the virtual interface the VPN creates.

Most traffic gets sent through a VPN only because of a default gateway (set by the VPN) in the client's routing table. If the client's ISP were to have their DHCP server set one or more specific routes that are broad enough to cover most of the global address space, they would effectively override that default gateway. I believe that's the scenario described in the article.

Note that the "ISP" here could be a mobile operator, an internet cafe, an airport, someone running a wifi access point that looks like the airport's, or a guest on the same local network running an unauthorized DHCP server.