Linux generally presumes that you run trusted software, not some proprietary program that is approximately malware. If you want a "sandbox" run that program as a separate unprivileged user or use bubblewrap.
But this statement basically says: "Linux has no good permission controls for running software". The assumption is flawed. Trusting software is not a true/false thing.
Yes, you can use sandboxing tools, but how many people use them properly? How many usability bugs do they still have?
Because of this heritage Android uses user accounts instead of namespaces a.k.a. containers (a much newer and less mature concept) to isolate apps from each other.