[motohagiography]

Could this be useful for doing dynamic exercising of apps in emulators to profile what kind of data they share?

I'm wondering if potentially it could be a way to close the loop on training a deep reinforcement learning model to exercise clicking on app UI's and using Solitude data over the network as the state/reward cycle.

[IncRnd]

Could you explain? I think I missed something but am interested in your comment. Are you saying to use Solutide data to train a model that hopefully approaches Solitude-like decisions? What could such a model detect that Solitude could not?

[motohagiography]

It's about whether you can use a DRL model to click at the random app interfaces in a more directed way, and then see its effects in the network traffic via Solitude. You could then download apps and their updates en masse and get automatic privacy profiles of how they actually behave instead of their bullshit policy statements and broad permissions.

App security companies likely do variations on this today, but Solitude in-effect instruments a network sandbox for dynamic analysis of app data handling. I'm saying blackboxing the app with more sophisticated tools like Solitude on the exercising and collection points could be more effective in determining its real behavior than trying to exercise these privacy related code paths directly.

I'm sure someone on this thread has a company that does this by now. The difference is the model clicking on the front end, and the sophistication of Solitude on the network output.

[IncRnd]

Gotcha, I understand now what you mean, train a model to click. That will be valuable for more than just Solitude on the back end. Thank you.

[no-dr-onboard]

Sounds like you could already do that by proxying the emulator through burp or running adb logcat with a grep filter. Is there something more than that that you need?