一句话：面向受监管软件开发，关注用生成式AI检测软件制品与监管文档之间的不完整、不一致和差异（DIID）。
在医疗、汽车、航空航天等高风险受监管领域，软件开发必须遵循行业、国家或国际监管标准。这些标准常由多份文档构成，范围从具体需求到通用指南；开发者需要证明、监管机构需要核验软件制品与监管文档之间的完整性和一致性。检测监管文档相对软件制品的不完整性、不一致性和差异（DIID）对双方都很重要，因为它可能指示潜在的不符合项或监管风险。标题表明论文采用生成式AI来检测DIID，但摘要未提供具体方法、数据集或评估结果等细节。
Software development activities must adhere to industry, national, and/or international regulatory standards to ensure security, safety, and other important properties in high-stakes, regulated domains, such as healthcare, automotive, and aerospace. These software standards are often codified via multiple documents, ranging from specific requirements to general guidance. Software developers must demonstrate—and regulatory authorities must then verify—the completeness and consistency between the software artifacts and the software regulatory documents. Detecting regulatory document incompleteness, inconsistencies, and/or discrepancies (DIID) with respect to software artifacts is important for both developers and regulators because they indicate potential non-conformance and/or software regulatory risk.
@inproceedings{paperbot9052,
title = {Generative AI Detection of Document Incompleteness, Inconsistencies, and Discrepancies},
author = {John Robert and Carlos Olea and Yash Hindka and Nanette Brown},
booktitle = {FSE},
year = {2025},
doi = {10.1145/3696630.3728724}
}