California has published the results of a statewide public-engagement project asking residents how artificial intelligence should be built, used and governed. The report offers a window into public priorities, but it is not a vote, a law or a statistically representative opinion poll. Its value—and its limits—are clearest when the numbers, recruitment process and policy status are read together.
What happened
On Oct. 13, 2026, Governor Gavin Newsom’s office announced the results of California’s first statewide use of the Engaged California digital-democracy program for artificial-intelligence policy. The official announcement describes the project as a way for Californians to help shape decisions about how AI is developed, deployed and regulated, and says the findings will help inform future policy work according to the governor’s office.
The project, titled Impact of AI on work, began in May with an open online survey available to Californians. The official report says 2,702 people completed or participated in that survey. A smaller group then took part in two-hour, facilitated Zoom discussions. The final report records 103 discussion participants across 14 conversations, from which staff grouped the input into 26 policy ideas in the Engaged California report.
The figures describe different stages of participation, not one single sample. The 2,702 figure is the broad listening phase. The 103 figure is the number who attended live discussions. The report also says 937 people were invited to those discussions, meaning the invite list was larger than the group that ultimately attended.
ABC10’s report independently summarizes the same three headline concerns and notes that residents suggested a wide range of possible policies, including stronger oversight, an AI tax, limits on AI in some fields, workforce training, parental controls and fees for employers that eliminate jobs because of AI ABC10 reported. Those ideas are proposals raised through the engagement, not enacted measures.
What participants said
The report’s three leading themes were selected because they appeared most often and attracted the strongest support in the live discussions. They are broad policy directions rather than draft bills.
Accountability for harms
The top theme was holding AI companies accountable for harm caused by their systems. Examples in the report include strengthening criminal liability and directing more resources toward enforcement. Participants also raised questions about who should be responsible when an AI output causes damage: the company that develops a model, the organization that deploys it, a human decision-maker, or some combination.
This concern does not establish that a particular company has caused a particular harm, nor does it resolve the legal question of liability. It records a public preference for clearer responsibility. In ABC10’s coverage, UC Berkeley experts Andrew Reddie and David Evan Harris framed the issue as a balance between AI’s benefits and risks, while debating where accountability should sit and whether everyday harms deserve as much attention as catastrophic scenarios as quoted by ABC10.
Resource demands and environmental effects
The second theme was protecting people and the environment from the resource demands of AI. The report points to data-center noise, pollution, water use and electricity use, as well as the need to study environmental impacts. The concerns connect technical infrastructure to local questions: how facilities affect nearby communities, who pays for added infrastructure, and whether increased demand could affect utility costs.
These are concerns raised by participants, not findings that the engagement independently measured data-center emissions, water consumption or household bills. The report does not claim that every facility has the same impact. It shows that resource use was central enough to become one of the three highest-ranked themes.
Rules for safe use
The third theme was setting rules for where and how AI can be used safely. Examples include requiring a human to make the final judgment in decisions affecting people’s lives and banning automated AI in hiring. Other ideas in the report include rules for personal data, labeling AI use, opt-out rights, education, parental controls and protections for workers affected by automation.
The breadth of these suggestions matters. The discussions were not simply for or against AI. Some participants proposed limits in high-stakes settings; others suggested using AI for work people do not want or cannot do. The official report says participants disagreed about how far regulation should go, how proposals should be funded and how rules should be enforced according to the project’s account.
How the process worked
California used two stages. First, anyone who found the open survey could respond. Second, interested survey-takers could be invited to live discussions. The state says it used a process called sortition to create a balanced invitee pool: demographic goals were based on California population estimates, and people who fit those goals were selected with a lottery-like method the methods page explains.
The targets included geography, age, race or ethnicity, gender and field of work. The method was intended to avoid hearing only from people who responded first and to produce discussion groups that roughly reflected the state’s diversity. That is a meaningful design choice, but it should not be confused with probability sampling of all Californians.
In the discussions, participants proposed ideas and voted as groups. Staff included ideas that received at least one vote in a discussion, combined similar proposals and summarized them into 26 policy ideas. The ranking then considered two measures: how often an idea appeared across the 14 discussions and how much support it received when it appeared. The report says frequency ranged from one to eight discussions and support counts ranged from zero to five.
The ranking therefore indicates which ideas were repeatedly discussed and comparatively supported in these sessions. It is not a referendum on the merits of a proposal. The report itself cautions that the ordering is not absolute and that the ideas are separate from existing policy or legislation.
What the methodology means
The strongest caveat is that the open survey was self-selected. People chose whether to find the survey and respond; they were not drawn through a probability sample in which every Californian had a known chance of selection. The 2,702 respondents may therefore differ from the statewide population in interest, experience, geography or views about AI. Their answers should be read as public input from participants, not as a statistically precise estimate of what all Californians believe.
The live-discussion group was designed to be more balanced than the open survey, but it was still drawn from people who had responded to the survey and expressed interest in further participation. Of 937 invitees, 103 attended. That creates a further participation filter: people who accepted the invitation and joined a discussion may not represent either all survey respondents or the broader public.
There is also a difference between demographic balance and representativeness of opinion. Matching targets for region or age can improve the diversity of a deliberative group; it cannot guarantee that the group’s policy preferences mirror the state’s preferences. The report is best understood as structured civic deliberation and idea generation, not a statewide mandate.
How AI was used in the project
The state’s methods disclosure is notable because the project was about AI while also using AI in its analysis. Staff used AI to convert discussion audio into transcripts, produce high-level summaries of discussions and survey responses, and tag portions of conversations by the ideas they covered. The state says this work ran in a secure environment, that transcripts and survey responses were not shared with AI vendors, and that the content was not used to train AI models according to the official methods disclosure.
The disclosure also draws a boundary around automated assistance. AI did not select discussion invitees, and the state says AI did not replace human judgment or write or edit participant words. Staff selected the themes and policy ideas, checked AI-generated statements against exact citations, and verified quotations against transcripts. The state also says it compared AI summaries with notes made by human notetakers and found they covered the same ground without inventing content.
Those safeguards reduce, but do not eliminate, questions about interpretation. Transcription can mishear speech; summaries can compress context; tagging choices can influence what is easy to count. The important point is that California has disclosed where AI entered the workflow and where officials say humans retained responsibility.
What the report does—and does not—change
The report is an input document. It does not itself create an AI tax, impose a ban on automated hiring, establish criminal liability or require a human decision-maker in every high-stakes setting. The administration says the feedback will help inform future policy decisions, while ABC10 explicitly reported that the report does not enact new regulations in its coverage.
That distinction also separates plans from outcomes. California may use the themes in future legislation, regulation, procurement or public programs, but the engagement alone does not show that any recommendation will be adopted. Conversely, an idea’s appearance in the report does not mean the state endorses it. The ranking is a record of what participants discussed and supported within this process.
For policymakers, the report supplies a map of concerns that can be tested against evidence, existing law, constitutional limits, costs and implementation capacity. For readers, it supplies a more modest conclusion: people who participated wanted a continuing role in decisions about AI and emphasized accountability, resource impacts and safety.
FAQs
Does the report represent all Californians?
No. The 2,702-person survey was open and self-selected, so it is not a probability sample and cannot produce a statistically representative statewide estimate. The 103 discussion participants were recruited from interested survey-takers and selected using demographic targets and a lottery-like process. That improves balance in the deliberations but does not turn the findings into a statewide vote.
What were the three main priorities?
Participants’ top themes were holding AI companies accountable for harms, protecting people and the environment from AI’s resource demands, and setting rules for where and how AI can be used safely. The themes were ranked using frequency across discussions and support within discussions, not by a statewide referendum.
What are some specific ideas participants raised?
The report includes examples such as stronger enforcement and liability, limits on data-center noise and water or electricity use, human final judgment in life-affecting decisions, restrictions on automated hiring, privacy rules, AI labeling, worker reskilling and easier opt-outs. ABC10 also reported suggestions including an AI tax, parental controls and fees for employers that eliminate jobs because of AI. These remain participant proposals unless adopted separately through law or policy.
Did AI decide which people were invited?
The state says no. Invitees were selected through the disclosed sortition process, with demographic goals and random selection within those goals. AI was used for transcription, summaries and tagging, including an assessment of respondents’ outlooks to help create a mix of viewpoints, but staff retained responsibility for the invitee process and reviewed the analysis.
Does publication of the report create new regulation?
No. The report records public input and may inform future decisions. It does not itself enact a regulation, allocate funding or impose liability. Any binding measure would need to come through the applicable legislative, regulatory or administrative process.
Conclusion
California’s exercise is significant less because it settles AI policy than because it documents an attempt to bring residents into a difficult policy conversation. Its headline numbers show reach and participation; its methodology shows an effort to structure deliberation; its caveats show why the results should not be treated as a mandate. The practical test now is whether officials use the themes transparently—explaining which ideas advance, which do not, and why.