With the rise of AI-generated code, programming courses face new challenges in detecting code plagiarism. Traditional methods struggle against obfuscation techniques that modify code structure through statement insertion and deletion. To address this, we propose a novel approach based on tolerant token matching designed to enhance resilience against such attacks.We evaluate our method through three experiments on a real-life dataset with AI-obfuscated plagiarisms. The results show that our approach increased the median similarity gap between originals and plagiarisms by 1 to 6 percentage points.