The rise of AI coding tools like GitHub Copilot has been a game-changer, boosting productivity and making coding more accessible. But this power comes with a potential downside: misuse in education and the risk of buggy or insecure code. How can we tell if code was written by a human or an AI? That's the challenge tackled by researchers in a new paper exploring AI-generated code detection. Existing methods for spotting AI-generated text often fall short with code because of its structured nature. Code relies on specific keywords and syntax rules, resulting in many predictable "low-entropy" tokens. These methods, which work well for the nuances of human language, struggle with the more rigid structure of programming languages. The researchers made a key observation: when AI rewrites code it *itself* generated, the changes are minimal. However, when rewriting human-written code, the AI tends to make more significant alterations. This insight led to a clever solution: use the AI's rewriting behavior as a test. Their method involves having an AI rewrite a given code snippet and then comparing the original and rewritten versions. A high similarity suggests the original code was AI-generated. This "zero-shot" approach, meaning it doesn't need prior training on labeled examples of AI and human code, has shown promising results. In tests, it significantly outperformed existing methods, raising hopes for a reliable way to detect AI-generated code. This research has important implications for education, ensuring fair assessments, and for software development, where identifying AI-generated code can help prevent vulnerabilities. While this new method is a significant step forward, the ongoing evolution of AI models means the cat-and-mouse game of detection continues. Future research will need to address increasingly sophisticated AI code generation techniques and potential countermeasures to this detection method.