Mia Bonta — Mia Bonta is repeatedly acknowledged during the hearing opening and early remarks, with speakers identifying her as Assembly Member/Chair Bonta and thanking her for her leadership, opening comments, and role in the discussion.
Mia Bonta — Addressed as an Assembly Member in the greeting.
Craig Poitowski — Introduced as the Cedars-Sinai presenter giving an overview of generative AI in health care.
Dr. Daniel Yang — Dr. Daniel Yang introduced himself as a practicing internal medicine physician and vice president of AI in emerging technology at Kaiser Permanente, then expanded on his role leading the enterprise responsible AI program and framed the discussion around the benefits and risks of AI.
Fawad Bud — Fawad Bud introduced himself as founder and CEO of Penguin AI and drew on his experience in healthcare data leadership to argue that U.S. healthcare is burdened by a major supply-demand mismatch and roughly a trillion dollars a year in administrative waste. He said prior authorizations, claims, coding, billing, and other paper-heavy workflows delay care and create a strong opportunity for generative AI to rewire healthcare, especially in administrative, back-office, and front-office tasks. He described AI use cases such as summarizing documents, verifying coverage, extracting codes, reviewing denials, and speeding prior authorization decisions from days to under an hour, while noting that clinical AI use cases are higher risk and require much stricter evaluation. He also emphasized that AI systems must be built with strong governance, compliance, privacy protections, and bias detection, and warned that healthcare data is biased and often excludes safety-net communities. Bud argued that California is well positioned to lead, but that equitable access requires including safety-net populations in training data with proper consent and oversight, since model developers who sidestep Medicaid/Medi-Cal data risk increasing bias against those communities.
Rebecca Bauer-Kahan — Rebecca Bauer-Kahan is repeatedly referenced as the chair during the hearing’s opening segment. She is identified as chair, thanked for her leadership and work on the Privacy Committee, and invited to give opening remarks/comments.
Rebecca Bauer-Kahan — Thanked as Chair Bauer-Kahan.
Daniel Yang — Referenced as having discussed the supply-and-demand mismatch in health care.
Amy McDonough — Amy McDonough introduced herself as leading Google’s global health solutions team and described her career at the intersection of health and technology, including work at Fitbit and Google Health. She explained that Google aims to build health into products and services people already use, and framed the broader healthcare context by noting workforce shortages, rising chronic conditions, and administrative burdens. She then began outlining Google’s four key approaches, starting with providing accessible health information through Search and YouTube.
Mia Bonta — Thanked as Chair Bonta.
Darshana Patel — Dr. Patel opened the question period and asked the witness panel about model performance, focusing on language adaptation and natural language processing. She followed up by asking how the models handle accents, cultural names, and culturally specific phrasing, referencing translation errors in Google voice messages as an example.
Darshana Patel — Dr. Patel asks the final question before the panel moves on.
Darshana Patel — The speaker responds to a question addressed to Dr. Patel about liability and where safety incentives fall in the AI ecosystem.
Dr. Yang — Named as one of the presenters being questioned about model performance.
Dr. Yang — Dr. Yang discusses how the AI tool and similar predictive algorithms learn from real-time clinical data and physician observations, then addresses concerns about whether such systems are being monitored for racial disparities and could worsen outcomes for Black women. He further explains that predictive algorithms, such as those used to identify patients likely to miss appointments, can reflect social barriers like transportation, work schedules, and child care, and that the same algorithm can be used in different ways depending on how its output is applied—either to worsen access through practices like double-booking or to improve care through outreach.
Mr. Butt — Named as one of the presenters being questioned about model performance.
Mr. Butt — Mr. Butt is referenced throughout a discussion on AI oversight and policy risks. The speaker thanks him for raising concerns about bias in AI systems, cites his earlier comments on bias, agrees with his call for strong governance, compliance, bias detection, and testing, refers to his remarks about plans and algorithms determining approvals, and finally asks for his view on where AI is headed and what policy pitfalls are likely.
KP representative — The KP representative explained that quality assurance testing identified a problem where the system could miss text when language switched, prompting vendor fixes and provider education about not using the tool as an interpreter or translation system. They clarified that other systems are used for translation, and that providers are trained on the technology’s strengths and limitations. The representative also noted that the scribe technology has been used in over six million encounters and is continuously monitored through provider feedback.
Fawad Butt — Fawad Butt discussed how AI technologies can mis-handle names, noting that his own last name is often censored, but said he expects pronunciation and name recognition to improve significantly in the next 12 to 18 months. He then broadened the point to warn that model performance can still be uneven if systems are trained on commercial and Medicare data without Medi-Cal data, which can disadvantage underserved populations and still require human review for inference and medication-unit errors.
Penguin AI representative — Said human review is required before key decisions are made, even if AI reduces the time needed.
Unidentified physician speaker — The physician speaker explained that AI outputs are reviewed by a physician and should be judged against the human standard rather than perfection. They emphasized that the AI-generated notes and the C-section prediction tool are meant to supplement clinical information, not replace clinician judgment or automatically drive decisions. The speaker said the tool can help clinicians weigh options such as watchful waiting versus proceeding with delivery, and noted that it is tested using real-world data across diverse patient populations and runs in the background as an added source of information.
Unidentified legislator or committee member — The legislator asked about the applications of the maternal health tools, including preeclampsia screening and a C-section prediction model, and questioned whether the reported 90% accuracy reflects the tool itself or how physicians use it. They raised concern that such tools could lead to more C-sections, which carry greater risk than vaginal births, and emphasized the importance of avoiding unintended harm. The legislator then thanked Mr. Butt for discussing bias in these systems and asked how the tool was tested to ensure good outcomes for women of color, noting broader maternal mortality disparities affecting Black women and other women of color.
Dr. Oktowski — Was addressed by the speaker regarding early risk preeclampsia screening and C-section prediction.
OB team and informatics team — They are described as reviewing the tool and making adjustments as needed, though no official longitudinal study has been done.
Ms. McDonough — The speaker asks for her view on where AI is going and the likely policy pitfalls.
Kara Carter — Kara Carter is introduced as the first panelist for the second panel, identified as being from the California Health Care Foundation. She then introduces herself as Senior Vice President for Programs and Strategy at the foundation and begins framing her remarks on AI in the safety net.
El Sol Neighborhood Educational Center — Cited as a community-based organization using AI tools to support community health work.
Dr. Zied Opermeyer — Dr. Zied Opermeyer is introduced as the next panelist and then begins his testimony as a physician and researcher at Berkeley, explaining that AI done right can save lives, lower costs, and expand access to care.
Rebecca Bauer-Kahan — The speaker referenced her point that AI learns from human-created data and that those data have flaws.
Rebecca Bauer-Kahan — Addressed as an Assembly Member in the greeting.
Jasmeet Bains — The speaker referenced the Attorney General's office in California as part of efforts to hold AI accountable. The transcript says 'Rob Banta,' which appears to be a transcription error.
Dr. Michelle Mello — Dr. Michelle Mello testifies that AI in hospitals is usually not a cost saver because it creates ongoing monitoring and staffing costs, and that the main value is often relieving overwork and capacity constraints rather than reducing expenses. She argues that developers often shield themselves with liability and warranty disclaimers, leaving hospitals and physicians likely to bear responsibility for errors, which creates weak incentives for safety. She emphasizes that AI tools require meaningful monitoring beyond simply checking whether they are turned on, noting that human oversight can erode as users trust the system and stop reviewing outputs, even when tools like ambient scribes only save modest amounts of time. She then supports a stronger state role in AI governance, comparing it to the Common Rule and institutional review boards in human subjects research, and concludes that regulation should focus on how AI is used downstream rather than restricting access to data or dictating tool design.
Dr. Sharp Collins — Asks about federal impacts on AI governance, culturally competent and unbiased AI, and whether federal data use may change under the current administration.
Dr. Obermeyer — Is invited to address the question about federal data and the current administration; the response discusses Section 1557 and algorithmic bias enforcement.
questioning legislator (unnamed) — The legislator raises concerns about a phone-based heart monitoring device, comparing it to Life360 and questioning its privacy implications. They then emphasize that any discussion of such technology should be framed in the context of data privacy, balancing privacy concerns against the potential life-saving benefits of health monitoring.
Dr. Mello — Speaker turns to Dr. Mello to discuss the developer-deployer dynamic.
Dr. Mello — Speaker asks Dr. Mello about applying IRB protocols to patient data use.
Dr. Patel — Speaker addresses Dr. Patel and then Dr. Mello with a question about IRB protocols and patient data.
Dr. Overmeyer — The speaker references Dr. Overmeyer in connection with prior remarks about using AI in safety-net communities, then asks Dr. Overmeyer to elaborate on the application of AI for those communities, including concerns about equity, bias, and policy recommendations related to Medi-Cal.
Ms. Carter — Ms. Carter is referenced in a discussion about the use of AI for safety-net communities, with follow-up prompting for her perspective on equity, bias, and policy recommendations, including implications for Medi-Cal and related public programs.
Chris Nielsen — Chris Nielsen, education director for the California Nurses Association, testifies about the risks of AI in nursing. He warns that while AI is often promoted as a solution to staffing shortages, it may instead worsen understaffing, displace jobs, undermine clinical judgment, and introduce bias and high error rates.
California Nurses Association — The witness argues generative AI can fail to save time and can increase nurses' workload by requiring review and correction of outputs. The witness calls for strong regulation, worker participation in deployment decisions, a precautionary framework, and the right of workers to override or object to new technologies.
Dr. Brent Sugimoto — A family physician and HIV provider from Lifelong Medical Family Medicine Residency describes benefits of ambient scribing, but says safety-net and rural settings cannot afford such tools. He warns about transcription bias, the need for representative data, workforce training, and leadership to incentivize primary care AI development.
Brenton Hill — Represents the Coalition for Health AI (CHAI) and introduces the organization as a large multi-sector alliance focused on responsible, trustworthy health AI.
AB2013 — AB 2013 is referenced as an example of legislation supporting AI transparency and disclosure standards. The speaker commends the Assembly for passing it and urges enforceable vendor disclosure requirements so the transparency is meaningful, then later cites it again as a framework that may prove effective in 2026.
David Ford — Introduces himself as CEO of CMA Physician Services, a subsidiary of the California Medical Association.
Dr. Sugimoto — Referenced as having described an AI hallucination involving an AIDS patient.
Christine von Rasefeld — Christine von Rasefeld, a board member of the Light Collective and co-author of the AI Rights Initiative, testifies in support of ethical AI governance in health care. She emphasizes the need for patient-led governance, transparency, strong privacy protections, and clear recourse provisions to ensure AI is developed and deployed ethically and equitably.
Mr. Ford — Addressed as one of the witnesses being asked about infrastructure support in safety-net settings.
Mr. Hill — Addressed as one of the witnesses being asked about infrastructure support in safety-net settings.