Security products
Collection of various thoughts on building and maintaining security products to solve problems.
Building a security product
What should you think about when building a security product? The original content was located...somewhere...but I did not persist the source. I will attempt to find it later, since the content sounds like it's from someone at Google.
- You should build large-scale AI systems to detect and block threats at scale before those threats can reach users. Examples of this working are safe browsing, where the user is stopped from reaching a malicious site, or identifying phishing emails before the hit the inbox. Pure rules-based approaches can't scale to stop all of those threats.
- Keeping up with constantly evolving attacks requires continuously improving and retraining detection systems. Otherwise, the adversaries eventually can subvert whatever decision boundary is in place. This concept is the Red Queen Hypothesis, where you need to adapt and evolve to survive since your adversaries are doing the same thing.
- What do you do if the situation is borderline? Give the user as much context as possible and rely on them to make the final decision. Choices like a warning banner will let them know to be aware, but without additional details, they may ignore it. If there are things they need to research on their own, are there ways that you could make their research easier?
- Over time, there will be more targeted attacks against journalists, hacktivists, politicians and campaigns, executives, fintech users, and celebrities. Can you make security products that work in those cases without getting in the way of their job or life?
I had saved some of these notes months before joining Duo Security, but the underlying concepts ring really true to what the team has been doing and what they are trying to get done.
Universal design
The Coalesce 2021 talk "Inclusive design and dbt" primarily
discussed how dbt and SQL are good cases of universal design in
data products. Universal design is the idea that designing something
to meet the accessibility needs of a few can improve the use for all. A
real-world example are sidewalk cutouts at intersections: designed for
wheelchair use, but helpful for a variety of people and situations.
Additionally, if you don't take the time to intentionally include
people, you will unintentionally exclude people.
Universal design is characterized by
- Equitable Use:
- Flexible Use:
- Simple and Intuitive Design:
- Perceptible Information: communicate information effectively to the user
- Error Tolerance:
- Low Effort: facilitate efficient and comfortable use (e.g. reduce boilerplate, minimize repetitive actions)
- Proportional Size in Space:
When considering the problem space of what a problem is trying to solve, I have in the past focused on how I would want to solve it versus what would solve the issue in the best way. That can be an issue when someone else needs to use the product. Without taking the time to see what the needs of the users are or how they are using it, the design of the product will not be as good as it could be.
For security products, this becomes more important as we are trying to solve a security problem for the users in a way that makes sense. An overly verbose product with intractable domain-specific information is not as effective as something that explains or guides the user through how they should work with the system. Instead of reporting an explanation code, you can provide a readblae description.
Programming languages have improved how their error messages show up to improve how new coders work with the language, but that also improves how everyone works with the language.
Created: December 5, 2021