> ## Content Index > Fetch the complete content index at: https://www.thedelatorrereview.com/llms.txt > Use this file to discover other available public pages before exploring further. # California Defines AI: AB 2885 and Its Impact - URL: https://www.thedelatorrereview.com/california-defines-ai-ab-2885-and-its-impact/ - Published: 2025-01-01T11:09:00.000Z - Updated: 2026-08-28T00:51:35.000Z - Description: California’s AB 2885 establishes a technology-neutral definition of artificial intelligence focused on AI systems, autonomy, inference, and outputs. This article explains the definition, how AI systems differ from AI models, where the definition applies, and what it means for AI governance. - Author: Lydia - Tags: Large Language Model (LLM), AI model, AI system, California, Automated Decision-Making Technology (ADMT), AI Governance, Algorithmic Accountability, Generative AI, Artificial Intelligence (AI) > **Key Takeaways: (1)California defines AI broadly** as a system that infers from inputs how to generate outputs. **(2) The definition is technology-neutral** and covers more than generative AI. **(3) AI systems can have varying levels of autonomy** and may involve human oversight. **(4) AI systems and AI models are different concepts**; California’s definition focuses on systems. **(5) Context and use matter** when determining how AI laws apply. **(6)Automated and high-risk decision systems are separate concepts** specific to the state-agency framework. **(7) There is no universal California AI definition**; always check the applicable law. --- **In-Force Date: January 1, 2025** In January 2025, California enacted a statutory definition of artificial intelligence (AI) that provides an important starting point for understanding the state's rapidly developing AI regulatory framework. It also provided a definition for *"high-risk automated decision system."* In 2024, Governor Gavin Newsom signed [**Assembly Bill 2885 (AB 2885)**](https://leginfo.legislature.ca.gov/faces/billTextClient.xhtml?bill%5Fid=202320240AB2885&ref=thedelatorrereview.com), authored by Assembly-member [Rebecca Bauer-Kahan](https://en.wikipedia.org/wiki/Rebecca%5FBauer-Kahan?ref=thedelatorrereview.com). The law establishes a common definition of AI for several provisions of California law. --- > **The California definition of AI systems is broad, technology-neutral, and focused principally on** **what a system does rather than on a particular technology or technical architecture**. It also closely reflects an emerging approach to defining AI around concepts such as autonomy, inference, objectives, inputs, and outputs. --- For organizations developing, deploying, procuring, or evaluating AI systems in California, understanding this definition is increasingly important. Before determining whether a particular AI-related obligation applies, organizations first need to understand whether the technology at issue falls within the relevant statutory definition. However, [AB 2885 ](https://leginfo.legislature.ca.gov/faces/billTextClient.xhtml?bill%5Fid=202320240AB2885&ref=thedelatorrereview.com)is primarily a **definitions and harmonization measure**. It should not be read as a comprehensive California AI governance law. As opposed to the [EU AI Act](https://www.thedelatorrereview.com/the-eu-ai-act/), it does not prohibit particular AI systems; impose general obligations on every developer or deployer of AI; establish a universal AI risk-management framework for California businesses; require every business to conduct an AI impact assessment; or make every use of AI a high-risk automated decision. ## **The California Definition of Artificial Intelligence** AB 2885 amended **California Government Code § 11546.45.5** to define artificial intelligence as: > **an engineered or machine-based system that varies in its level of autonomy and that can, for explicit or implicit objectives, infer from the input it receives how to generate outputs that can influence physical or virtual environments.** It is important to note that [AB 2885](https://leginfo.legislature.ca.gov/faces/billTextClient.xhtml?bill%5Fid=202320240AB2885&ref=thedelatorrereview.com) establishes a common definition of AI systems ONLY for the statutory provisions it amended. That does **not necessarily mean every California statute or regulation using the term “artificial intelligence” will cross-reference to this definition**. Specifically, it applies across the following existing statutory contexts: - **State agency automated decision systems** **(California Government AI and Automated Decision Systems -** Government Code §§ 11545-11548)— Requires the California Department of Technology to inventory high-risk automated decision systems proposed, developed, procured, or used by covered state agencies. These are systems that assist or replace human decision-making in areas with significant effects, including housing, education, employment, credit, health care, and criminal justice. The AI definition and the definitions of automated decision systems, and high-risk automated decision system([Cal. Gov. Code § 11546.45.5(a)-(c)](https://leginfo.legislature.ca.gov/faces/codes%5FdisplaySection.xhtml?lawCode=GOV§ionNum=11546.45.5.&ref=thedelatorrereview.com))matters because AI is one of the technologies that can qualify as an automated decision system. The inventory must address the system’s decisions, data used, benefits and alternatives, and safeguards addressing cybersecurity, privacy, accuracy, discrimination, and bias. - **Social media terms-of-service reporting** — The is California’s Content Moderation Requirements for Internet Terms of Service law ([Cal. Bus. & Prof. Code §§ 22676–22677](https://leginfo.legislature.ca.gov/faces/codes%5FdisplayText.xhtml?lawCode=BPC&division=8.&title=&part=&chapter=22.8.&article=&ref=thedelatorrereview.com)) requiring covered social media companies to publish specified terms of service and submit semiannual reports to the Attorney General describing their content-moderation practices and results. The definition of AI ([Cal. Bus. & Prof. Code § 22675(b)](https://leginfo.legislature.ca.gov/faces/codes%5FdisplaySection.xhtml?lawCode=BPC§ionNum=22675.&ref=thedelatorrereview.com).) matters because those reports must distinguish content flagged or actioned by “artificial intelligence software” from content handled by employees, contractors, community moderators, civil-society partners, or users ([Cal. Bus. & Prof. Code § 22677(a)(5)(B)(iv)–(v)](https://leginfo.legislature.ca.gov/faces/codes%5FdisplaySection.xhtml?lawCode=BPC§ionNum=22677.&ref=thedelatorrereview.com).) Accordingly, the statutory AI definition determines which systems count as AI for these reporting purposes. - **California Online Community College** — The California Online Community College Act ([Cal. Educ. Code §§ 75000-75013)](https://leginfo.legislature.ca.gov/faces/codes%5FdisplayText.xhtml?lawCode=EDC&division=7.&title=3.&part=46.5.&chapter=&article=&ref=thedelatorrereview.com) establishes and governs California’s public online community college, including its administration, educational programs, student support, technology, accessibility, funding, and accountability. It specifically directs the college’s Research and Development Unit to use technologies including machine learning and artificial intelligence to develop student-support tools, such as a virtual help desk available at any time. The definition of AI ([Cal. Educ. Code § 75002(a)](https://leginfo.legislature.ca.gov/faces/codes%5FdisplaySection.xhtml?lawCode=EDC§ionNum=75002.&ref=thedelatorrereview.com).) is relevant because it determines what technologies qualify as AI when the Act directs the college to use AI in developing these student-support systems. - **Warehouse economic development subsidy reporting** — The law requires local agencies granting economic development subsidies to warehouse distribution centers to publicly disclose information about the subsidy, jobs, wages, benefits, and other employment impacts. Cal. Gov. Code § 53083.1(a)–(g). The AI definition ([Cal. Gov. Code § 53083.1(j)(1)](https://leginfo.legislature.ca.gov/faces/codes%5FdisplaySection.xhtml?lawCode=GOV§ionNum=53083.1.&ref=thedelatorrereview.com),) matters because annual reports must disclose any known net job loss or replacement caused by automation, AI, or other technologies. --- > **A Common Definition Does Not Mean a Universal California AI Definition** --- ### California Is Consolidating Around This Definition AB 2885 did more than harmonize the definition of AI across the statutes it amended. Since its enactment, California lawmakers have continued using the same technology-neutral definition in new AI legislation addressing very different subjects. For example: - **AI training-data transparency:** California’s Artificial Intelligence Training Data Transparency Act uses the same general AI definition as the foundation for its narrower definition of generative AI and related disclosure requirements. [**Read more about this law.**](https://www.thedelatorrereview.com/californias-ai-training-data-transparency-law-what-generative-ai-developers-must-disclose/) - **AI voice disclosures:** California’s robocall law uses the definition to determine when a voice generated or significantly altered by AI triggers a disclosure requirement. [**Read more about this law.**](https://www.thedelatorrereview.com/californias-ai-voice-disclosure-rule-when-robocalls-must-tell-you-the-voice-is-artificial/) - **AI civil liability:** Civil Code § 1714.46 uses the same definition in its rule preventing defendants from arguing that AI autonomy itself caused the alleged harm. [**Read more about this law.**](https://www.thedelatorrereview.com/californias-ai-liability-rule-why-the-ai-did-it-is-no-longer-a-defense/) These laws regulate very different activities, but they increasingly begin with the same answer to the threshold question: **What counts as AI?** --- > California is **consolidating around a common AI vocabulary while regulating particular uses of AI through context-specific laws.** --- ## Breaking Down California’s Definition of Artificial Intelligence The definition therefore contains several important concepts: 1. an **engineered or machine-based system**; 2. that **varies in its level of autonomy**; 3. operates in pursuit of **explicit or implicit objectives**; 4. **infers from the input it receives** how to generate outputs; and 5. generates outputs capable of **influencing physical or virtual environments**. Each of these elements helps define the boundaries of what qualifies as an AI system under California law and is examined individually below. ### 1\. An “Engineered or Machine-Based System” California does not define AI by reference to a particular technical method such as machine learning, neural networks, large language models, or generative AI. This makes the definition **technology-neutral**. - A system therefore does not necessarily need to use a particular type of model or algorithm to potentially qualify as AI. - Conversely, merely marketing a product as “AI-powered” does not determine whether it falls within the statutory definition. --- > **Practice Tip: Do not rely on product labels when conducting an AI inventory.** An organization's AI governance process should evaluate how a system actually functions rather than simply asking whether a vendor or internal development team describes it as “AI.” --- **California’s definition focuses on an AI “system,” not AI models**.[ Government Code § 11546.45.5](https://leginfo.legislature.ca.gov/faces/codes%5FdisplaySection.xhtml?lawCode=GOV§ionNum=11546.45.5.&ref=thedelatorrereview.com) defines AI as an “engineered or machine-based system” that can infer from inputs how to generate outputs capable of influencing physical or virtual environments. - An **AI model** is the computational component that has been trained or otherwise developed to recognize patterns, make predictions, generate content, classify information, or perform other AI-related functions. A large language model (LLM), for example, may receive text as an input and predict or generate text as an output. - An **AI system** is broader. It is the operational technology in which one or more models may be incorporated together with other components—such as software, interfaces, databases, prompts, business rules, APIs, and human processes—to perform a particular function. --- > A useful way to think about the distinction between AI models and AI systems is: **AI model = the underlying computational capability; AI system = the model + other technical and operational components used to accomplish a particular objective** --- For example, a LLMs may be incorporated into a customer-service application. The model generates responses based on the inputs it receives. The AI system may include that model together with the customer interface, retrieval database, system prompts, authentication controls, escalation rules, logging, and mechanisms for transferring certain questions to a human representative. --- > **Practice Tip: Inventory AI systems and use cases, not just AI models.** Organizations should understand **where the model is deployed, what other components surround it, what inputs the system receives, what outputs it produces, and how those outputs are used**. The same model can be incorporated into multiple AI systems with very different purposes, risks, and legal implications. --- > An effective inventory should ideally identify at least: **Model or technology → system or application → intended use → inputs → outputs → decision or action influenced → affected individuals → applicable legal requirements.** --- ### 2\. AI Can Have Different Levels of Autonomy California's definition does not require a system to operate entirely independently of humans. Instead, **AI systems may operate along a spectrum:** - At one end, a system may generate recommendations that are reviewed and acted upon by a person. - At the other, a system may perform tasks or generate outputs with substantially less human involvement. --- > The existence of **human review or human involvement does not necessarily mean that a system falls outside the definition of AI**. --- Autonomy is a characteristic of AI that can varies within a spectrum—not an all-or-nothing threshold. ### 3\. Explicit or Implicit Objectives California does not limit AI to systems that have been given a formally stated goal. Both systems that have explicit and implicit objectives may qualify: - An **explicit objective** may be directly programmed or specified—for example, identifying spam, recommending content, predicting demand, or classifying an image. - An **implicit objective** may arise from the system's design, optimization process, training, or operation even where the objective is not expressed as a simple programmed instruction. ### 4\. The Importance of “Inference” One of the most significant elements of the definition is that the system must be capable of **inferring from the input it receives how to generate outputs**. This language distinguishes AI from conventional software that simply executes predetermined instructions. Traditional software may follow rules such as: **If X occurs → perform Y.** An AI system may instead use inputs to determine, predict, classify, recommend, generate, or otherwise produce an output based on patterns or relationships represented within the system. The statutory focus is therefore not on whether software processes information. It is on whether the system can **infer from its inputs how to generate its outputs**. A calculator, for example, processes an input and produces an output. But processing inputs according to fixed mathematical rules is different from a system that infers from its inputs how an output should be generated. --- > **Practice Tip:** When determining whether software falls within California's definition of AI, ask **how the output is produced**. A system that merely executes predetermined instructions may present a different analysis from one that uses inputs to infer classifications, predictions, recommendations, decisions, or generated content. --- ### 5\. AI Outputs Can Take Many Forms California does not limit AI to systems that generate text, images, audio, or other content. This can potentially encompass many types of outputs, including: - predictions; - recommendations; - classifications; - scores; - generated text; - images, audio, or video; - decisions or decision support; - commands affecting physical devices; and - actions affecting digital environments. This is one reason the California definition should not be understood as a definition of generative AI. --- > **Generative AI is one category of AI technology. California's statutory definition is broader.** --- ## What Is a “High-Risk Automated Decision System”? For purposes of the **state-agency inventory requirement**, [Government Code § 11546.45.5 ](https://leginfo.legislature.ca.gov/faces/codes%5FdisplaySection.xhtml?lawCode=GOV§ionNum=11546.45.5.&ref=thedelatorrereview.com)defines a **high-risk automated decision system** as an automated decision system used to assist or replace human discretionary decisions having a **legal or similarly significant effect**. This includes decisions materially affecting access to or approval for: - housing or accommodations; - education; - employment; - credit; - health care; and - criminal justice. Under this provision, an **AI system may qualify as an automated decision system, which may in turn qualify as a high-risk automated decision system depending on how it is used.** Importantly, this hierarchy is specific to the state-agency inventory framework in § 11546.45.5\. The definitions of “automated decision system” and “high-risk automated decision system” do **not** apply to the other statutes that AB 2885 amended to incorporate California’s general definition of AI. Within this framework, classification depends not simply on the underlying technology, but on **how the system is used and the significance of the decisions it supports or makes**. --- > “AI” and “Automated Decision Systems” Are Not the Same Thing**. Artificial intelligence is a general technological concept. An automated decision system is defined by a more specific use of computational technology in decision-making that materially impacts people.** --- ### Conclusion [AB 2885](https://leginfo.legislature.ca.gov/faces/billTextClient.xhtml?bill%5Fid=202320240AB2885&ref=thedelatorrereview.com) provides an important building block for California’s developing AI framework by establishing a common definition of AI across several statutes. But **defining a technology as AI is only the beginning of the legal analysis**. Organizations must also consider how the system is used, the decisions or activities it supports, and the definitions and requirements of each applicable California law. ## Legal Citation **California Assembly Bill 2885 (2023–2024), Chapter 843**, approved September 28, 2024. Relevant provisions include: - **Cal. Gov. Code § 11546.45.5** — definition of artificial intelligence, automated decision system, and high-risk automated decision system. - **Cal. Bus. & Prof. Code § 22675** — incorporates the definition for specified social-media provisions. - **Cal. Educ. Code § 75002** — incorporates the definition for the California Online Community College. - **Cal. Gov. Code § 53083.1** — incorporates the definition for specified warehouse economic-development-subsidy reporting requirements. #### Related California AI Laws - **California’s AI Training Data Transparency Law: What Generative AI Developers Must Disclose** — How the definition operates in California’s generative AI training-data transparency regime. [**Read the article**](https://www.thedelatorrereview.com/californias-ai-training-data-transparency-law-what-generative-ai-developers-must-disclose/) - **California’s AI Liability Rule: Why “The AI Did It” Is No Longer a Defense** — How the definition is used in California’s emerging approach to AI and civil liability. **Read the article** - **California’s AI Voice Disclosure Rule: When Robocalls Must Tell You the Voice Is Artificial** — How the definition has been incorporated into California’s existing robocall framework. **Read the article** --- **End Note: A Now-Repealed Deepfake Provision** \- AB 2885 also incorporated the AI definition into **Government Code § 11547.5**, which addressed **deepfakes, digital content forgery, and digital content provenance** and required the Secretary of Government Operations to develop a coordinated plan for addressing these technologies. **That provision was temporary and was repealed on January 1, 2025.** Former Cal. Gov. Code § 11547.5(a), (e). ![](https://storage.ghost.io/c/54/ef/54efeb65-4f2f-479e-b4bb-a9ee526feeaa/content/images/2026/08/Screenshot-2026-07-04-at-4.45.19---PM-13.jpeg)