Public postsecondary education: generative artificial intelligence systems: procurement standards: training.

The required composition of the intersegmental working group was narrowed by removing 'faculty' from the list of advocacy organizations. The text now requires 'representatives from student advocacy organizations' instead of 'representatives from student and faculty advocacy organizations.' This slightly reduces the mandated diversity of stakeholder input on GenAI procurement standards and training.

Section 66098.1 no longer conditions the working group and training obligations on the act of 'providing a GenAI system to students, faculty, or staff.' The segments must now convene the working group and provide training as standing duties regardless of whether a specific GenAI system has been deployed.

The separate provision requesting that the University of California join the working group (former subparagraph (C)) was removed as a standalone item, though UC's participation is still referenced in the broader section language.

Systemwide academic senates were added to the list of required participants in the intersegmental working group, giving faculty governance bodies a direct seat at the table for developing procurement standards and training.

The requirement in Section 66098.2 that an updated training be provided to students, faculty, or staff before their continued use of a product that has undergone a material change in functionality or privacy practices was deleted. Segments must still review and update training annually, but there is no longer an explicit trigger requiring retraining upon material changes.

Section 66098.3 removed the specific start date of January 1, 2027 for the interim reporting requirement. The obligation now simply applies 'until GenAI procurement standards are adopted,' with no defined earliest start date, making the window potentially shorter or more flexible.

The legislative reporting requirement in Section 66098.3 was changed from a report submitted BEFORE executing a procurement contract to a report submitted WITHIN 60 DAYS AFTER executing a systemwide contract. This shifts the obligation from a pre-approval gate to a post-hoc accountability mechanism.

The privacy training requirement was reduced from 'a disclosure of the privacy policies applicable to a GenAI system, including how user data and prompts may be collected, stored, or used to train the system' to 'an overview of the privacy policies applicable to GenAI systems, and internet website links to the privacy policies.' This significantly lowers the depth of information students, faculty, and staff must receive.

The working group is now described as 'intersegmental' and segments must 'jointly convene and participate in' it, rather than simply 'convene a joint working group.' This emphasizes equal participation across all three higher education segments.

Labor organizations that represent employees of the three segments were added as required members of the intersegmental working group, giving employee unions a formal seat in developing GenAI procurement standards and training.

Responsibility for California Community Colleges obligations was narrowed from 'the California Community Colleges' broadly to 'the office of the Chancellor of the California Community Colleges,' specifying a particular administrative office rather than the entire system.

The term 'generative artificial intelligence product' was replaced throughout with 'generative artificial intelligence system,' changing the legal characterization of what is being regulated.

References to 'institutional policy' were changed to 'systemwide policy' in multiple places, clarifying that the relevant policies are at the system level (CC, CSU, UC) rather than individual campus level.

The bill's target was narrowed from any 'artificial intelligence tool, program, or application' to specifically 'a generative artificial intelligence product (GenAI) system,' excluding non-generative AI tools from its requirements.

The legislative intent statement ('It is the intent of the Legislature to enact future legislation relating to artificial intelligence in postsecondary education') was deleted, removing a forward-looking signal about future AI legislation.

New definitions were added for 'artificial intelligence' and 'generative artificial intelligence' (both cross-referencing Government Code Section 11549.64) and a new term 'sycophancy,' defined as the tendency of a GenAI system to overly agree with or flatter a user even when they express factually inaccurate or illegal ideas, for the purpose of optimizing engagement.

A joint working group must be convened by the CCC and CSU Chancellors (with UC requested to join) to develop procurement standards and training. The group must include privacy experts, student and faculty advocacy representatives, community-based organizations, civil society groups, and academic researchers. Recommendations are due to system leaders by January 1, 2028.

Procurement standards must ensure GenAI systems meet nine minimum protections: documented risk assessment for harms and bias; no output of harmful content (nonconsensual intimate imagery, suicide ideation, disordered eating, unlicensed mental health therapy); no excessive sycophancy prioritizing engagement over accuracy; safety monitoring with human escalation for self-harm or harm to others; transparent incident reporting; written data use and deletion policies; data minimization with regular audits; vendor disclosure of all training data sources including copyrighted content and student-generated content; and vendor compliance with privacy/civil liberties and labor standards.

Training must cover institutional GenAI policies including academic integrity, privacy disclosures (how user data and prompts are collected, stored, or used for training), and guidance on system limitations including potential for inaccurate or misleading outputs. Each segment may develop distinct versions tailored to students, faculty, and staff.

New Section 66098.2 requires institutions to review training at least once per academic year, update it when policies, laws, or system functionality/privacy practices change, provide updated training before continued use after material changes, and maintain records of completed trainings (name, date, version) for a minimum of four years.

New Section 66098.3 requires CCC and CSU (and requests UC) to submit a written report to the Legislature and relevant policy committees before executing any GenAI procurement contract during the period from January 1, 2027 until procurement standards are adopted. The report must describe the system, vendor, intended use, affected population, evaluation and selection process, planned training, and vendor data practices.

The fiscal committee designation changed from 'no' to 'yes,' indicating the amended bill now carries fiscal implications requiring review by the legislative fiscal committee.

New Section 66098 requires California Community Colleges and California State University to provide training to students, faculty, or staff before providing any AI tool, program, or application. Training must cover relevant AI use policies and a disclosure of the product's privacy policies.

The University of California is requested (not required) to provide the same training before deploying AI products to students, faculty, or staff.

Scope narrowed from all postsecondary education (including private institutions) to public postsecondary education only: California Community Colleges, CSU, and UC.

New Article 14 added to Chapter 2 of Part 40 of Division 5 of Title 3 of the Education Code, titled 'Artificial Intelligence,' creating a dedicated statutory framework for AI in public postsecondary education.

Fiscal committee review changed from 'no' to 'yes,' indicating the amendment triggers a fiscal impact analysis.