HB0823 would require developers of generative artificial intelligence systems to publicly disclose training-data documentation on their websites before releasing, or substantially modifying, a covered system. The required documentation is extensive and includes the sources or owners of the training data, how the data supports the system’s purpose, dataset size or estimates, labeling characteristics, whether the data are public domain or protected by copyright or other IP rights, whether the data were purchased or licensed, whether personal or aggregate consumer information is included, whether the data were cleaned or modified, the time period of collection, when the data were first used in development, and whether synthetic data generation was used.
The bill defines “developer” broadly to include private persons and state or local government units that design, code, produce, or substantially modify a generative AI system. It applies to generative AI systems released on or after January 1, 2022 for use by the general public in Maryland, whether or not users pay for access. It excludes systems made exclusively for certain specialized purposes, including hospital medical staff use, affiliate use, physical safety, protection of confidential personal information, fraud detection or prosecution, aircraft operation, and federal national security, military, or defense uses. The bill takes effect October 1, 2025, and would amend Maryland’s State Finance and Procurement article by adding a new transparency section and updating existing AI definitions.
Its main legal impact would be to create a new state-level transparency obligation for generative AI developers operating in Maryland, expanding the state’s existing artificial intelligence framework. By requiring public disclosure of training-data provenance and characteristics, the bill would affect AI vendors, deployers, and potentially public-sector developers, while also creating a compliance expectation tied to product release and later substantial modifications. The measure does not ban generative AI systems or regulate model performance directly; instead, it focuses on disclosure and documentation.
The available legislative context shows no recorded votes or committee testimony, and the hearing was canceled, so there is little direct evidence of formal support or opposition in the record provided. Based on the bill’s structure, the general policy sentiment appears to favor transparency, accountability, and informed public scrutiny of AI systems, especially regarding data sourcing and privacy. At the same time, the breadth of the disclosure requirements suggests potential concern from developers about administrative burden, proprietary information, and the practicality of documenting large or dynamic training datasets.
The most likely points of contention are the scope of covered systems, the level of detail required for public disclosure, and whether the transparency mandate could expose trade secrets or create compliance costs that are difficult for smaller developers or public entities to meet. Exemptions for healthcare, safety, fraud prevention, aviation, and national security uses indicate an effort to limit the bill’s reach to general-purpose consumer-facing systems, but those carveouts may also raise questions about where the line should be drawn between public transparency and operational confidentiality.
HB0823 would amend Maryland’s State Finance and Procurement law to add a new section requiring public training-data disclosures for generative artificial intelligence systems. It would define “generative artificial intelligence,” “developer,” and “substantially modified,” and would require covered developers to publish detailed documentation about the datasets used to train a system before release or major updates. The bill would affect AI developers, including private companies and state or local government units, while exempting several specialized systems used for healthcare, safety, fraud detection, aviation, and federal defense purposes.
The bill’s apparent policy direction is favorable toward transparency and accountability in artificial intelligence, with no recorded vote history or committee testimony available in the provided materials. Because the hearing was canceled and no votes are listed, there is no formal evidence of opposition or support in the record beyond the bill’s text. The structure of the bill suggests a generally pro-regulation, pro-disclosure sentiment aimed at making generative AI development more visible to the public and policymakers.
Likely areas of contention include the breadth and specificity of the required disclosures, the administrative burden on developers, and whether public posting of training-data details could reveal proprietary or sensitive information. Developers may object to the cost and feasibility of tracking and summarizing large or continuously changing datasets, while supporters are likely to emphasize transparency, privacy, and accountability. The exemptions for certain high-stakes or sensitive uses suggest an attempt to balance those concerns, but they may also be debated as either too narrow or too broad.