The widespread adoption of Large Language Models (LLMs) has accelerated significant progress in code generation while simultaneously giving rise to concerns related to academic honesty, software security, and the attribution of authorship. Traditional detection tools often fail when faced with diverse programming languages and modern LLMs. We introduce GPAD (Generative-Process-Aware Detection), an extension of CodeBERT that augments the attention mechanism with token-level entropy signals and stylometric features, improving robustness across generators and domains. Our Solution is designed for the SemEval-2026 Task 13 on detecting machine-generated code under diverse conditions by evaluating generalization to unseen languages, generator families, and code application scenarios.