3 ms·

> Aside: anyone who wants to talk about doing data analysis without access to the underlying data is cordially invited to explain how to do this to statistician

> Aside: anyone who wants to talk about doing data analysis without access to the underlying data is cordially invited to explain how to do this to statisticians who only understand R and local Jupyter notebooks. Provision some isolated machines with access to production data, and let the data scientists login remotely. You'll be able to tell who is accessing the data and when, and won't need to install invasive bossware on their systems. It's a fairly straightforward solution. It's one thing to control access to sensitive data. Its another to use that to justify monitoring employees all day, even when they aren't accessing sensitive data.

Kalium 4y ago
Now you need to convince them to use this system. You also need to convince them to not use this system, where they're injecting arbitrary code into the sandbox remote machine, to download the data to their local machines where they can work in the way they understand and are comfortable with. Inevitably some person will react badly to the perceived lack of trust or object to what they see as making their job more difficult for no reason. There are a lot of options for monitoring for that kind of thing. You monitor the remote machine, you monitor the network traffic between it and their local terminal... and the most valuable will be monitoring their local machine to associate with the rest of the access and usage information. Among other things, you need to be able to distinguish between someone using the data scientist's credentials and the data scientist using their own credentials. Your idea is a wonderful approach! It improves security, privacy, traceability, and auditability. It may not be the same as removing the need for endpoint monitoring, unfortunately.
- asdff 4y ago
  You can never perfect endpoint monitoring. At the very least, the engineer could see data, go to the bathroom, and write down what they saw on a scrap of paper. As long as eyeballs see data this is possible no mater what software you install on device, or even if you have an armed guard sitting in the office right behind the engineer.

Kalium 4y ago

Now you need to convince them to use this system. You also need to convince them to not use this system, where they're injecting arbitrary code into the sandbox remote machine, to download the data to their local machines where they can work in the way they understand and are comfortable with. Inevitably some person will react badly to the perceived lack of trust or object to what they see as making their job more difficult for no reason. There are a lot of options for monitoring for that kind of thing. You monitor the remote machine, you monitor the network traffic between it and their local terminal... and the most valuable will be monitoring their local machine to associate with the rest of the access and usage information. Among other things, you need to be able to distinguish between someone using the data scientist's credentials and the data scientist using their own credentials. Your idea is a wonderful approach! It improves security, privacy, traceability, and auditability. It may not be the same as removing the need for endpoint monitoring, unfortunately.

asdff 4y ago
You can never perfect endpoint monitoring. At the very least, the engineer could see data, go to the bathroom, and write down what they saw on a scrap of paper. As long as eyeballs see data this is possible no mater what software you install on device, or even if you have an armed guard sitting in the office right behind the engineer.