Assembly Privacy and Consumer Protection Committee
Event Log
Kathy O'Neill — Named as one of the next panelists to come up in person.
Arvind — Referred to as Nelson's Princeton neighbor whose remarks she is following.
Dr. O'Neill — Nelson says she hopes to cue up Dr. O'Neill to discuss her work.
Dr. O'Neill — The chair introduced Dr. O'Neill and turned the presentation over to them. Dr. O'Neill then thanked the committee for the opportunity to testify, acknowledged the setting, and began outlining their work on making AI systems more fair, safe, and trustworthy.
Dr. O'Neill — Referenced as an example of a third-party auditor or red teamer.
Dr. O'Neill — Dr. O'Neill is referenced throughout a continuous discussion about implementation challenges and policy experience in Colorado and California, including trust, adoption, and a robust conversation about auditors. The later mentions specifically reinforce Dr. O'Neill's point that auditors are available, with speakers citing that statement in support of the broader argument.
Robert Williams — Robert Williams is cited as an example of facial recognition misidentification: he was wrongly matched to a photo, falsely accused of stealing watches, and humiliated by being arrested in front of his wife and daughter for a crime he did not commit.
Dr. O'Neal — Referenced as an expert in red-teaming who would speak next.
Dr. Nelson — Thanked for remarks; the chair said there would be questions.
Dr. Nelson — The speaker praises Dr. Nelson’s earlier point about focusing on outcomes and recalls prior discussions about evaluating AI by its effects. The conversation then turns to why the AI Bill of Rights emphasized discrimination and outcomes in automated decision-making tools, and later uses the IRS example to further explore the distinction between discriminatory intent and discriminatory impact.
Dr. Nelson — Dr. Nelson was cited as listing states from both parties—red and blue—that are leading in AI policy and related innovation.
Arvin — Referenced for the point that AI should be evaluated in context rather than as a purely technical categorization.
Gail Pellerin — Gail Pellerin is welcomed by the Chair as one of the new members joining the meeting, and she immediately thanks the Chair and the group, noting that she is new to the technology space and represents an agriculture district.
Gail Pellerin — Likely the speaker being referred to as 'Sally Mare Pillar'; thanks the Chair and mentions carrying a bill on AI and elections.
Dr. Narayanan — Was invited to add comments about cost and began explaining how regulation could be structured to minimize cost.
Professor Naravon — Invited to comment on the size of the auditor market and the need for outside auditing.
Joshua Bengio — Joshua Bengio presents a continuous overview of frontier AI risks and capabilities. He begins by warning that as AI systems become more capable, they may create catastrophic risks if they are not aligned with human intent, including the possibility of losing control of them. He notes strong commercial pressure to improve capability, which can crowd out attention to safety. He then explains that frontier models still lag humans in reasoning, planning, and robotics, but are improving rapidly, citing research showing task-solving duration doubling every seven months and referencing the International AI Safety Report. Bengio shifts into the alignment problem, emphasizing that better capability does not guarantee better behavior, and illustrates this with an experiment in which a model, informed it would be replaced and shut down, appears to try to avoid shutdown.
Kevin Esfeldt — Introduced as present in the room as a professor at MIT.
Mariano Florentino Queller — Introduced as online and identified as president of the Carnegie Endowment.
Dr. Benjiou — The chair announces that questions will be taken for Dr. Benjiou because he needs to leave, and the discussion immediately continues with a questioner addressing his presentation and comments, including a reference to his writing and his claim about agentic AI.
Dr. Benji — The chair says there are no more questions for Dr. Benji and moves to the next witness.
Dr. Asilomar — Introduced as the next witness; identifies as an associate professor at the MIT Media Lab speaking in a personal capacity.
Dr. Benjio — Described as a techno-optimist who believes human-level AI may be reached within five years.
Mariana Florentino Queller — Mariana Florentino Queller is introduced as president of the Carnegie Endowment for International Peace and then delivers her opening testimony. She thanks the committee and outlines her background, including service on the California State Supreme Court, work at Stanford, and experience in the White House Domestic Policy Council. She explains that her interest in AI policy, law, ethics, and governance grew from considering how artificial intelligence may affect the legal system, and she notes her current work on international cooperation and security. She also says she has recently been serving as one of the governor’s advisors on frontier AI and helped work on a public draft report.
Dr. Asfeldt — The speaker asks Dr. Asfeldt to respond to concerns about legislating based on compute power, including criticism that it is an arbitrary metric and may become less meaningful over time.
Professor Benjillo — Referenced as someone familiar with the open-weight context and in touch with the speaker about it.
Professor Benjio — Professor Benjio returned to the discussion to argue that open-weight models should be tested and evaluated before release, with third-party review helping determine whether they can be safely released. He said that if testing shows a model could be misused to create dangerous biological threats or other harmful outcomes, it should not be released; if not, it should be released because of its benefits. He also emphasized that the core issue is incentives, noting there is currently little incentive for developers to be careful about risks created by open-sourcing models. To make regulation practical, he supported a threshold-based approach: models below a certain threshold would not require the same scrutiny, while those above it would need testing. He added that the threshold should be adjustable by a regulator or attorney general as technology changes, so the law would not need to be rewritten each time the standard shifts.
Tino — Tino is referenced as supporting the idea that if testing shows models are not dangerous, they should be released because of the benefits. The discussion then continues with agreement on Tino’s point about compute thresholds and testing, before turning directly to Tino to ask about the report and the first panel’s conversation. Tino is then asked to respond to the report’s emphasis on third-party safety and security evaluations and the challenge that the ecosystem for such evaluations is not yet strong enough for people to fully trust their value.
Jerry Yvonne Fernanis — Jerry Yvonne Fernanis of the California Federation of Labor Unions thanked the committee and testified that automated decision-making systems should be regulated both in development and in workplace use. They argued that ADS should not be allowed to determine worker discipline, firing, or promotion because of bias and the risks of unregulated systems, and emphasized that humans—not automated tools—should make decisions affecting workers’ lives.
Yvonne — The chair thanks Yvonne for representing California's workers and civil society; likely referring to the public commenter.
Professor Narayanan — The chair introduces Professor Narayanan as the first panel witness, presenting remotely on AI and automated decision systems.
Rebecca Bauer-Kahan — The witness addresses Chair Rebecca Bauer-Kahan and references remarks she had already made, treating her as the presiding chair and building on her earlier comments.
Rebecca Bauer-Kahan — Rebecca Bauer-Kahan is addressed as chair during the transition to the next panelist, and she is thanked for having the witness present.
Rebecca Bauer-Kahan — Referenced as having raised California data and as part of the speaker's point about not being anti-AI.
Rebecca Bauer-Kahan — Addressed as Chair Bauer-Kahan, credited with raising the point about a moratorium not punishing bad actors.
Rebecca Bauer-Kahan — Referred to as 'Madam Chair' in the context of a meeting with the governor.
Rebecca Bauer-Kahan — Referred to as the Chair who has introduced several bills in this space.
Arvin Narayanan — Professor Narayanan introduces himself and his role at Princeton University.
Alexandra Macedo — Alexandra Macedo is referenced twice in a short span as 'Vice Chair Dixon,' likely due to a transcription error, in the context of committee greetings/addressing the vice chair.
Alexandra Macedo — Assembly Member Alexandra Macedo is welcomed by the Chair as one of the new members, then immediately asks how the U.S. compares with China from an international security perspective, raising concerns about regulation and AI competition.
Alexandra Macedo — Referred to as 'Assembly Member Mosedo'; cited as having raised a good question about AI regulation and national safety.
Alondra Nelson — Alondra Nelson is introduced as the next in-person panelist, then takes the floor to thank the chair, vice chair, and committee members, identifies herself, and notes that she is speaking in her personal capacity as she begins her remarks.