Get off my Input! Input Validation in the Era of Zero Trust and AI

Protecting applications at the first line of defense.

Description

This talk in its core is a plea for input validation. Because my thesis is: The security of programs must (also) be defended at the inputs!

Almost all programs can be reduced to a black box that processes inputs and delivers outputs. In this model, attacks are based on the fact that I as the attacker send inputs that lead to unplanned and/or unwanted effects (e.g. the code that I have just sent is executed). Which inputs are allowed in my program is the core question of input validation. However, a shortcut is often taken here by assuming that the inputs come from a benevolent player and are within the desired range.

Against the backdrop of an era of zero trust architectures, however, we want to implement software so securely and resiliently that it does not require any presupposed security guarantees. We will only achieve this if we finally take input validation seriously and implement it. We should not forget the current trend in artificial intelligence, which in some cases requires a completely new type of input validation (keywords: prompt injection, model poisoning)

The talk starts in the first part with the main thesis and formulates the major challenges that zero trust and AI pose for input validation. We introduce the input-output model and use a few examples to reduce software and vulnerabilities to the problem of input validation. The second part of the talk then deals with measures on a syntactic and semantic level for input validation, their feasibility and their effectiveness.

(Disclaimer: Input validation is of course only a tool and I am absolutely aware that input validation alone is not enough).

Why the committee chose this talk

Protecting application becomes more important as we remove the security features of the network layer.