Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{187810,
        author = {Mr. Yatheendra KV and Dr. A M Sudhakara},
        title = {Robust Detection of Humanized AI Text: A Training Framework Using Prompt-Based AI Outputs and Paraphrased AI Content},
        journal = {International Journal of Innovative Research in Technology},
        year = {2025},
        volume = {12},
        number = {6},
        pages = {6763-6766},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=187810},
        abstract = {Large Language Models (LLMs) generate high-quality text that increasingly resembles human writing, making AI-generated content detection a challenging problem. Existing detectors often fail when models are prompted explicitly to “sound human” or when their outputs are paraphrased through third-party APIs. This paper proposes a robust training framework for AI-text detection based on systematically collecting (1) prompt-based adversarial AI content and (2) paraphrased AI content from multiple LLM APIs, and integrating these samples into a multi-task RoBERTa detection model. We introduce a standardized dataset construction pipeline that uses OpenAI/ChatGPT, Claude, Gemini, and LLaMA APIs to generate adversarially humanized text via targeted prompts and multiple decoding strategies. We further construct paraphrased-AI samples via back-translation APIs (DeepL, Google Cloud Translate), paraphrase tools (Quillbot API, T5/PEGASUS paraphrase APIs), and structured human-like rewrite prompts. Empirical experiments demonstrate that including adversarial and paraphrased AI samples during training significantly improves cross-model generalization, robustness to humanization attacks, and detection accuracy on unseen LLM outputs. This paper provides the full dataset-generation methodology, training pipeline, evaluation suite, ablation studies, and guidelines for deploying robust AI-generated text detection in academic and enterprise environments.},
        keywords = {},
        month = {November},
        }