Thread by @padresj

Not too long ago I was part of an exercise to determine how difficult it would be to identify people involved in protests w/just the metadata (including location) from the cellular network.

We had a semi-simulated data set that used real info w/swapped IDs

... It was CRAZY easy

Once we sorted the data, we could track movement patterns of the devices & see which devices were communicating & moving w/each other.

In our data set, the vast majority of devices were burners or otherwise obfuscated, but we quickly figured out which devices were daily drivers.

As it turns out, that's enough.

If you find the daily drivers (and that becomes STUPID easy if you have access to more metadata) then you have a real-ID reference point.

From there, you can spot all the devices that are moving or communicating w/the real-ID.

Even if they're burners, you can still identify them as part of the real-ID's social network.

From there, it's just a matter of looking at past or future connections to the real-ID.

Even w/o expansive past or future metadata, we were able to connect burners to real-IDs with ~50% accuracy by looking at when burners connected or d/c'd from the cell network & correlate it with when other real-IDs from the prime suspect's network did the opposite.

This wasn't groundbreaking work.

... but it really hammers home something that I think everybody needs to take to heart:

"In a connected world, your security is only as good as the OpSec practices of the people you allow in your circle."