California Enacts Additional Generative AI Bills Touching on Training Data and Healthcare Decisions

California Governor Gavin Newsom vetoed the controversial SB 1047, rejecting its broad AI safety mandates in favor of more targeted regulations. Instead, he signed 17 other bills addressing specific generative AI risks, signaling a shift from broad liability frameworks to precise, issue-focused compliance. This legislative pivot highlights the difficulty of regulating rapidly evolving technology through blunt instruments, favoring niche protections for consumer rights and safety over systemic model constraints. The centerpiece of this regulatory shift is AB 2013, which mandates transparency regarding training data. Developers must publish high-level summaries of the datasets used to train models available to Californians, disclosing sources, intellectual property status, and data types. This requirement directly challenges the industry norm of keeping training methodologies opaque for competitive advantage, aiming to facilitate accountability in ongoing copyright and privacy litigation by making data origins visible to the public. This development is critically relevant to the open data community as it establishes a legal precedent for disclosing training inputs. By forcing the revelation of data sources and characteristics, the law potentially unlocks critical information for researchers auditing model biases and verifying consent. While limited in scope to publicly available services, it sets a benchmark for transparency that could encourage broader open data practices, allowing civil society to scrutinize the foundational materials powering modern artificial intelligence systems.

Source: natlawreview.com
Published on 2024-10-01