OWASP Global AppSec US 2021

Protecting critical web APIs continues to be more challenging than ever, as attackers constantly learn to adapt to the evolution of the defenses that are in place on popular websites. Critical APIs in this context include login, account registration, password recovery, add to cart, checkout for most commerce, banking and gaming websites, but also seat / bed / stateroom availability check on travel and hospitality websites. In order to make the attack cost effective and efficient, attackers have long ago adopted the power of automation and built botnets that are capable of sending large numbers of requests at scale. For the most part, attackers have mastered the art of impersonating known good systems, like a MacBook running the latest versions of Google on recent OS X. They also have learned to spread their attack by taking advantage of the wide variety of proxy services available for a fee, in order to to make their activity less obvious. These common attack techniques can generally be detected using methods that take advantage of device and browser characteristics collected on the client side (device intelligence) or reputation systems (IP intelligence).

However, some attackers have become significantly more advanced and subtle and are able to send requests with clean fingerprints and IP addresses, making device and IP intelligence detection methods less effective. In this case, the next best thing in terms of detection is to check the user behavior to see how they interact with the website (i.e the path they took before interacting with the critical endpoint) as well as how they interact with the device that is sending the request (i.e mouse movements, key presses events, touch events, and coordinate changes from various sensors for mobile devices).

After reviewing the fraud threat landscape and how to defend against the most common attack methods, we’ll take a deep dive into behavioral detection methods. In particular, we’ll define what behavioral detection is, the signals that are worth collecting, methods of processing the information and finally building machine learning algorithms to effectively detect suspicious activity.