New AI Tool Catches a Sneaky Coding Mistake That Breaks Apps

Five researchers have built a tool called BugProbe that catches one of the sneakier mistakes in software: the swapped-argument bug. It happens when a programmer writes a command with two pieces of information in the wrong order — like a transfer slip that reads "send Alice 50 dollars" when you meant "send 50 dollars to Alice." The code often still runs, so the error can sit unnoticed until something breaks or money moves the wrong way.

To teach it, the team pulled 132,739 Python files from the 1,000 most popular projects on GitHub, a site where programmers share code. From those they generated 3.37 million practice examples. They also hand-checked 55 genuine swapped-argument bugs from real project histories, to test whether the tool catches mistakes that actually happened in the wild — not just ones a computer invented.

The clever part is what BugProbe does not need. Older detectors had to hunt down the original definition of a command to compare names. In Python that link is often impossible to trace, because Python never declares what kind of information each command expects. BugProbe instead reads clues around each command — how the values are used — and blends that with machine learning to judge whether the order looks wrong.

The catch: this is a research result, not a product you can install. It was accepted at an academic conference (SCAM 2026) and tested mainly on a small set of 55 verified bugs. It also only handles Python, and any detector like this risks false alarms that waste a developer's time. Still, the direction matters: software that catches its own mistakes before users ever feel them.