Data Flow Analysis

Data flow analysis is a way of tracing a single value's journey through a program from start to finish: where it is created, what it passes through, and where it ends up being used.

It's similar to following a river from its source to its mouth to check whether any pollutants slip in along the way. In code, this means tracking how a value entered by a user moves through one function after another and is eventually used on screen or in storage, watching for any point where a dangerous value flows through unfiltered.

This approach catches problems that are hard to find by looking at a single file alone. A value received in File A that gets passed to a function in File B and then executed as-is in File C is easy to miss when each file is examined separately, but tracing the value's flow all the way through reveals it.

The article explains that Claude Security's new model 'doesn't look at files one by one, but tracks the flow of data moving between files and infers how components interact with each other.' It's easy to mistake this for a brand-new technique invented by AI, but data flow analysis itself is a long-established code analysis method — what's new here is applying an AI model to carry out that analysis instead of a human.

You can try something similar to data flow analysis by pasting in a code snippet and asking:

"Trace, in order, which functions this user input passes through and where it ultimately ends up being used. Point out any spots where the value is passed along without validation."