>if our bad actors use VPN's to hide their traffic then it would become impossible for ISP's to see what they're doing or using
you just transfered a problem from ISP level to VPN operator level. While you could argue that using multiple VPNs from different countries could make this somewhat harder, the problem still exists. Especially if you consider metrics other than IP, for example specific packet sizes or timing patterns (for example, instead of users connecting to given IP, the adversary would look for users sending 640 byte packets every 300 seconds).
While the arguments that encryption of messages makes it impossible to know the contents of messages (and thus using the contents as evidence), however the ability to uncover the members/employees/cooperators of bad actor would make it easier to investigate them and/or use other means of targeted surveilance to obtain evidence. Also this would make it easier to infiltrate bad actor, since one of the uncovered users could be then coerced into cooperation.
(All above assumes that the app/platform is used only by members of "bad actor" and noone outside that organization is using the app. It is completely different if there are other users, perhaps even bad-actor users being a minority.)
With the developers outside jurisdiction, the problem is that while they of course might or might not be required to comply with the law, but they can still be coerced/manipulated/otherwise encouraged into providing a "patch" (backdoor) into the application.
I believe that much better solution would be to simply use any popular platform as a transport layer, with independent end-to-end encryption. Possibly with some steganography as well. The simplest example would be users exchanging memes/cat pictures - this will not stand out in any ISP/VPN traffic analysis. It will also not stand out (that much) in content analysis by any entity that can decrypt/access plain-content. The images being exchanged could then contain embedded (and end-to-end encrypted) content.
While this is still far from perfect - you could imagine detection of repetitive images being sent, content/timing patterns or actual analysis of attachments for steganography but all those still require significantly more resources to work on massive scale.
Alternative would be to use custom platform but having as many "external" (in a sense of not working with/for bad actor) users as possible