HB823 would require developers of generative artificial intelligence systems that are released for use by the general public in Maryland to publish documentation on their websites describing the data and datasets used to train those systems. The disclosure must be made by January 1, 2026, and again each time the system is released or substantially modified. The bill defines “generative artificial intelligence” broadly to include systems that generate synthetic text, images, video, or audio, and it defines “developer” to include private persons as well as state or local government units that design, code, produce, or substantially modify such systems.
The required documentation is detailed and includes the sources or owners of the data, how the data supports the system’s purpose, dataset size or estimates, whether labels were used, whether the data are public domain or protected by intellectual property 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, and whether synthetic data generation was used. The bill excludes certain systems made exclusively for hospital staff, affiliate use, physical safety, protecting confidential personal information, fraud or crime detection/prosecution, aircraft operation, and federal national security or defense purposes.
The bill would amend Maryland’s State Finance and Procurement Article by adding a new section on generative AI transparency and by adding a definition of “generative artificial intelligence” to the subtitle’s definitions. In practical terms, it creates a new state-law disclosure obligation for AI developers operating in Maryland, while leaving the underlying development and deployment of AI systems otherwise unchanged. It also extends the transparency requirement to public-sector developers, including state and local government units, if they develop or substantially modify covered systems.
Overall, the bill appears to reflect a policy preference for transparency and accountability in AI development, especially around training data provenance and data governance. Because there is no recorded committee transcript or vote history provided, there is no documented opposition or support in the materials supplied. Based on the bill text alone, likely points of concern would include compliance burden, protection of trade secrets or proprietary data, and the feasibility of disclosing detailed training-data information for large or frequently updated models.
HB823 would add a new disclosure regime to Maryland law for developers of generative AI systems, requiring public website documentation about training data before release or substantial modification. It would affect private AI developers and certain state or local government developers, while carving out several specialized uses. The bill would amend the State Finance and Procurement Article by adding § 3.5-807 and expanding the subtitle’s definitions, thereby creating a new statutory transparency obligation without directly regulating model outputs or banning any AI uses.
The available materials suggest a generally pro-transparency, pro-accountability approach to artificial intelligence. The bill’s sponsors are a bipartisan-looking group of delegates, and the text is framed as a disclosure measure rather than a restriction on AI deployment. However, because no committee transcript or vote record is provided, there is no direct evidence of debate, amendments, or recorded support/opposition in the supplied context.
The main likely points of contention are the breadth and specificity of the required disclosures, the administrative burden on developers, and whether the bill could force disclosure of proprietary, copyrighted, or otherwise sensitive training-data information. Another possible concern is the practicality of tracking and updating documentation for systems that are frequently retrained or substantially modified. Supporters would likely emphasize transparency, consumer awareness, and accountability in AI development, while critics may focus on compliance costs, trade secret concerns, and the difficulty of applying these requirements to complex AI systems.