Thread by @jeanqasaur

Just had an good conversation with @panmanphil about data flows:
🧐 Why enforcing privacy/access policies in software is hard
😫 Why static and dynamic analysis aren't the solutions of people's dreams
🛣️ Why data visibility/data mapping is The Way

THREAD 👇

@panmanphil 🧐 Enforcing security/privacy policies on data flows is not as simple as checking a rule as data leaves the database. *Which* rule you check for *what* data matters and is hard to keep track of. Either devs need to track this or software needs to track this. It's hard for devs.

@panmanphil 😫 Automatic policy enforcement requires automatic data tracking. Tracking data at runtime comes with unappealing performance overheads. Tracking it statically, before you run, gives you way too many false positives. Anything short of this leaves major gaps.

@panmanphil 😫 Also: people often use static/dynamic code analysis when what they really want is *software* analysis. Code analysis is going to give info on *one node at a time* of the service graph below. This graph is an ecosystem with a life of its own and we should treat it as such.

🛣️ Given that it is not (currently) possible to build a practical system to automatically enforce privacy/security rules, it is incredibly important to build tooling to make the rules getting enforced are the right ones. These tools should show what data is going where, when.

🛣️ We can't just apply existing program analysis to data mapping. They completely fall down because not only were they too conservative before, but they completely fall down in service-oriented architectures. This is why we've been working on a new approach at @AkitaSoftware.

@AkitaSoftware 💡 If you do find yourself in a position of needing to build a policy framework, however, when I was in grad school we did come up with a cute way of monkey-patching Django and using meta-programming in the ORM to programmatically stored policies. Happy to talk about that. 🥰