$ cat california-policy.md --date 2025-11-09 --read 12 min
last modified 2025-11-09

California's 2025 Tech Legislation: AI Transparency, Chatbots, and Privacy

California has enacted a suite of new legislation with a diverse set of impacts. Among other things, these bills are intended to:

  • Increase transparency in the process of building frontier AI models (defined later).
  • Ensure the well-being of individuals interacting with AI chatbot services.
  • Combat social media addiction.
  • Protect children and minors.
  • Prevent stalking via connected devices.

Most businesses operating in the State of California -- remotely or otherwise -- will likely need to implement compliance measures by the relevant deadlines to maintain their bottom line. Because California is the 4th largest economy in the world, this affects many businesses.

Much of the AI-specific legislation is aimed at large organizations with copious resources, so SMBs need not worry about those portions -- but the breach notification and chatbot rules apply to businesses of all sizes.

There's A LOT to cover here, so I'll be focusing on global impacts.

AI Legislation

It is important to note that California has been the USA's leader in sounding the horn on data, privacy, and AI legislation.

As of November 2025, AI's Black Box Problem remains unsolved: even the developers of large AI models cannot fully explain why a model produced a given output, because its behavior emerges from billions of learned parameters rather than legible, human-written rules. This makes model behavior difficult to audit, predict, or guarantee -- which is precisely why transparency legislation focuses on disclosure of processes rather than inspection of the models themselves.

SB 53: Transparency in Frontier Artificial Intelligence Act

Full Bill Text [3]: https://leginfo.legislature.ca.gov/faces/billNavClient.xhtml?bill_id=202520260SB53

Culminating a multi-year policy-making process, the ratification of the Transparency in Frontier Artificial Intelligence Act (TFAIA) [3] is perhaps the most noteworthy of the entire package. It's the first piece of comprehensive AI legislation regulating the development of frontier foundation models -- that is, cutting edge AI models -- in the U.S.

Last year, after vetoing a similar bill (California Senate Bill 1047 of 2024), Gov. Gavin Newsom tasked the Joint California Policy Working Group on AI Frontier Models to deliver a report "focused on artificial intelligence (AI) frontier models." [1][2] This report ultimately guided the creation of the finalized bill.

Fostering public trust in AI developers and service providers is the predominant goal:

  • It's meant to foster an environment of transparency surrounding the innovation and development of cutting edge, highly-capable foundation models.
  • Large frontier developers must write, publish, and actually follow a frontier AI framework describing how they incorporate national/international safety standards, assess catastrophic risk, and govern secure deployment. The framework must be reviewed and updated at least annually.
  • A transparency report must be published whenever a new frontier model (or a substantially modified version of one) is released -- covering its capabilities, intended uses, and restrictions.
  • Critical safety incidents must be reported to California's Office of Emergency Services within 15 days of discovery -- or within 24 hours if the incident poses an imminent risk of death or serious physical injury.
  • Whistleblower protections shield employees who report catastrophic-risk concerns, and large frontier developers must maintain an anonymous internal reporting channel.
  • Enforcement falls to the Attorney General, with civil penalties of up to $1 million per violation. Notably, there is no private right of action -- contrast this with SB 243 below.
  • The bill also establishes CalCompute, a consortium tasked with designing a public compute cluster to support safe, equitable AI research.

For context, the bill's section 1(f) [3] defines a foundation model as:

(f) "Foundation model" means an artificial intelligence model that is all of the following:

(1) Trained on a broad data set.

(2) Designed for generality of output.

(3) Adaptable to a wide range of distinctive tasks.

A short, high-level summary of the law can be gleaned from section 1(k) [3]:

"With the frontier of artificial intelligence rapidly evolving, there is a need for legislation to track the frontier of artificial intelligence research and alert policymakers and the public to serious risks and harms from the very most advanced artificial intelligence systems, while avoiding burdening smaller companies behind the frontier."

Large companies at the frontier of AI research are the primary focus, but a provision for smaller businesses is made with section 1(n) [3]:

"In the future, foundation models developed by smaller companies or that are behind the frontier may pose significant catastrophic risk, and additional legislation may be needed at that time."

The law applies only to organizations meeting these thresholds:

Compute Threshold: Definition of a "frontier model" -- section 2(i)

(i) (1) "Frontier model" means a foundation model that was trained using a quantity of computing power greater than 10^26 integer or floating-point operations.

(2) The quantity of computing power described in paragraph (1) shall include computing for the original training run and for any subsequent fine-tuning, reinforcement learning, or other material modifications the developer applies to a preceding foundation model.

Note: 10^26 operations = 100,000,000,000,000,000,000,000,000 operations

Revenue Threshold: Definition of a "large frontier developer" -- section 2(j)

(j) "Large frontier developer" means a frontier developer that together with its affiliates collectively had annual gross revenues in excess of five hundred million dollars ($500,000,000) in the preceding calendar year.

SB 243: Companion Chatbots

Full Bill Text [10]: https://leginfo.legislature.ca.gov/faces/billNavClient.xhtml?bill_id=202520260SB243

Senate Bill 243 focuses on the mental well-being of users of AI chatbots. It broadly shares many goals with TFAIA [3], attempting to set healthy precedents that increase the public's trust in AI services compliant with the policy.

It mandates the implementation of "…a protocol for preventing the production of suicidal ideation, suicide, or self-harm content to the user…" [10] into most AI chatbot services. Details of such protocols must be made available on service operators' websites. Additionally, it emphasizes the use of awareness notifications to prevent users from being misled into thinking an AI chatbot is a human.

Further provisions are made for "a user that the operator knows is a minor:"

  • It must be disclosed to the user that they are interacting with artificial intelligence.
  • Users must be clearly notified and reminded at least every three hours to take a break and "that the companion chatbot is artificially generated and not human."
  • "Reasonable measures" must be instituted to prevent the generation of sexually explicit material.

Annual Reports: Beginning July 1, 2027, operators [of AI chatbot services] must annually deliver a report on these efforts to California's Office of Suicide Prevention. Data from such reports must be de-identified. Data from the report must be made available on the operator's website.

Section 22603(a) -- The annual report shall consist solely of these data points [10]:

(1) The number of times the operator has issued a crisis service provider referral notification pursuant to Section 22602 in the preceding calendar year.

(2) Protocols put in place to detect, remove, and respond to instances of suicidal ideation by users.

(3) Protocols put in place to prohibit a companion chatbot response about suicidal ideation or actions with the user.

Importantly, it's crucial to understand what a "companion chatbot" IS -- and what it IS NOT.

Section 22601(b)(1) -- Definition of a "Companion chatbot" [10]:

(b) (1) "Companion chatbot" means an artificial intelligence system with a natural language interface that provides adaptive, human-like responses to user inputs and is capable of meeting a user's social needs, including by exhibiting anthropomorphic features and being able to sustain a relationship across multiple interactions…

Section 22601(b)(2) -- What isn't a "Companion chatbot" [10]:

(2) "Companion chatbot" does not include any of the following:

(A) A bot that is used only for customer service, a business' operational purposes, productivity and analysis related to source information, internal research, or technical assistance.

(B) A bot that is a feature of a video game and is limited to replies related to the video game that cannot discuss topics related to mental health, self-harm, sexually explicit conduct, or maintain a dialogue on other topics unrelated to the video game.

(C) A stand-alone consumer electronic device that functions as a speaker and voice command interface, acts as a voice-activated virtual assistant, and does not sustain a relationship across multiple interactions or generate outputs that are likely to elicit emotional responses in the user.

Notably, provisions are made for those suffering injury due to non-compliance (Section 22605):

A person who suffers injury in fact as a result of a violation of this chapter may bring a civil action to recover all of the following relief:

(a) Injunctive relief.

(b) Damages in an amount equal to the greater of actual damages or one thousand dollars ($1,000) per violation.

(c) Reasonable attorney's fees and costs.

Non-AI Legislation

SB 446: 30-day Data Breach Customer Notification Limit

Full Bill Text [9]: https://leginfo.legislature.ca.gov/faces/billNavClient.xhtml?bill_id=202520260SB446

This bill amends Section 1798.82 of the California Civil Code -- the state's existing breach notification statute. Section 1798 is most notably affiliated with the enactment of the California Consumer Privacy Act (CCPA) in 2018.

Currently, businesses operating in California are legally obligated to disclose a compromise of California residents' unencrypted personal information in "the most expedient time possible and without unreasonable delay…" [9].

Senate Bill 446 replaces the abstract time window, mandating (under normal circumstances) a maximum limit of 30 days between the first discovery OR notification of the compromise. In other words: organizations operating in California MUST notify affected parties within 30 days of the point at which the organization becomes aware a breach occurred.

Delaying a breach's disclosure is allowed in these cases:

  • "…to accommodate the legitimate needs of law enforcement…"
  • "…or as necessary to determine the scope of the breach and restore the reasonable integrity of the data system."

Section 1(d) [9] clearly lists the requirements of a sufficient breach notification:

…(1) The security breach notification shall be written in plain language, shall be titled "Notice of Data Breach," and shall present the information described in paragraph (2) under the following headings: "What Happened?" "What Information Was Involved?" "What We Are Doing," "What You Can Do," and "For More Information." Additional information may be provided as a supplement to the notice…

In addition to the quote above, section 1(d) contains many more granular requirements for breach disclosures. It also provides a pre-built template which can be used to fill out a written disclosure.

If a single breach results in a notice that must be issued to more than 500 California residents, a copy of this notice must be submitted to the Attorney General within 15 days of notifying affected consumers:

(f) An individual or business that is required to issue a security breach notification pursuant to this section to more than 500 California residents as a result of a single breach of the security system shall electronically submit a single sample copy of that security breach notification, excluding any personally identifiable information, to the Attorney General within 15 calendar days of notifying affected consumers of the security breach. A single sample copy of a security breach notification shall not be deemed to be within Article 1 (commencing with Section 7923.600) of Chapter 1 of Part 5 of Division 10 of Title 1 of the Government Code.

In practice, this may just result in organizations feigning ignorance until they're ready to disclose publicly -- or asserting that certain compromised data was already publicly available. That defense gets stronger the more personal information users leave exposed, which is one more argument for keeping your data away from the public gaze: scrubbing old posts across platforms, and being careful about where and with whom you share information.

AB 1043: Digital Age Assurance Act

Full Bill Text [4]: https://leginfo.legislature.ca.gov/faces/billNavClient.xhtml?bill_id=202520260AB1043

This bill adds a title to section 1798 of the California Civil Code (home of the CCPA), relating to consumer protection.

Beginning January 1, 2027, the law requires "operating system providers" to include a built-in user interface and application programming interface (API) to facilitate age verification within the operating system and other applications downloaded to the device. Covered application stores are also obligated to receive and act on these age-range signals.

California Civil Code Section 1798.500(g) [4] defines "operating system provider" as:

…a person or entity that develops, licenses, or controls the operating system software on a computer, mobile device, or any other general purpose computing device.

This broad definition raises complex compliance questions across the operating system ecosystem due to the diverse range of OS licensing -- most obviously within open source projects such as Linux.

Penalties are harsh with maximum penalties of $2,500 per child for each negligent violation and $7,500 per child for every intentional violation.

However, the bill does make provisions for providers making "good faith" compliance efforts:

California Civil Code Section 1798.503(b) [4]:

"An operating system provider or a covered application store that makes a good faith effort to comply with this title, taking into consideration available technology and any reasonable technical limitations or outages, shall not be liable for an erroneous signal indicating a user's age range or any conduct by a developer that receives a signal indicating a user's age range."

AB 566: California Opt Me Out Act

Full Bill Text [5]: https://leginfo.legislature.ca.gov/faces/billTextClient.xhtml?bill_id=202520260AB566

This bill adds section 1798.136 to the California Civil Code (home of the CCPA), relating to privacy.

The purpose of the bill is to make the process of opting-out of data collection, sharing, & selling more accessible.

Beginning January 1, 2027, businesses developing/maintaining a web browser MUST include user-configurable functionality to automatically send an "opt-out preference signal" to businesses interacted with via the browser [5].

It's worth mentioning that aside from Google Chrome -- the world's most popular browser -- the majority of modern web browsers already support this using Global Privacy Control (GPC) signals [6][7]. In this case, it appears the State of California is forcing Google's hand.

GPC is similar to the now-deprecated Do Not Track (DNT) browser feature, which usually went ignored by third parties receiving the signal. GPC now supersedes DNT.

Regulatory pushes behind California's CCPA and the EU's GDPR are largely responsible for the development of GPC. Pursuant to the CCPA, GPC communicates a "Do Not Sell or Share" request from a user's browser to third parties. It is likely that GPC will be utilized to implement this "opt-out preference signal" in Chrome and other browsers.

To privacy advocates, this is a positive stride. Though, without jurisdiction-dependent regulatory & transparency measures, whether parties receiving these signals will actually act in accordance with user opt-out preferences is ultimately unknown. In California, such measures are mandated under the CCPA.

Other Notable Legislation

The following is a brief run-down of all other legislation relevant to information security passed as of October 13th (closing day).

Inclusion of items within this list -- as opposed to full stories above -- is NOT meant to discount their impact. This legislation has been broad and all-encompassing, therefore I chose what I felt to be the most impactful.

AI

AI and Law

  • Targeted restriction of a specific type of legal defense argued in court cases involving AI (Assembly Bill 316) https://leginfo.legislature.ca.gov/faces/billTextClient.xhtml?bill_id=202520260AB316
    • Defendants having "developed, modified, or used" AI cannot assert "a defense that the artificial intelligence autonomously caused the harm to the plaintiff."
    • In simple terms: If the output of an AI agent causes harm to an individual (within the state of California), the service provider -- NOT the AI agent itself -- will be held directly responsible.

Non-AI (These directly mandate the inclusion of additional FEATURES in existing software/tech stacks)

Analyst's Comments

Taken together, this package raises real compliance burdens for large AI developers and sets new guardrails on chatbots, breach response, youth safety, and consumer privacy. Because of California's economic weight, expect spillover well beyond state lines: multi-state operators will likely just adopt the California baseline rather than fragment their products.

Key judgments

  • SB 53 concentrates its obligations on a small handful of large providers -- but don't expect the 10^26 compute and $500M revenue thresholds to sit still. As training practices shift (iterative fine-tuning, RL on top of prior models), the covered population could grow, and the bill's own text openly anticipates expanding to smaller developers later.
  • SB 243's private right of action ($1,000 per violation plus attorney's fees) is the sharpest enforcement tool in the package. Product teams should also watch the "companion chatbot" definition edges -- a productivity bot that picks up enough personality and memory could creep into covered territory without anyone deciding it should.
  • SB 446's hard 30-day clock will force incident-response playbooks to be rewritten. The law-enforcement and scope-determination exceptions remain, but organizations leaning on them should expect to document that reliance carefully.
  • AB 1043's OS-level mandate collides with how operating systems are actually governed -- especially open-source distributions -- and will likely need interpretive guidance before 2027. And while AB 566 mandates browser-side opt-out signals, whether downstream data receivers honor them still depends on enforcement.

Indicators to watch

  • Agency rulemaking clarifying SB 53's documentation expectations, and how "material modifications" fold into the compute threshold.
  • The first public operator reports under SB 243, and any enforcement actions or private suits tied to inadequate protocols.
  • Attorney General guidance or actions referencing SB 446's 30-day clock.
  • OS and browser vendor roadmaps for age signaling and opt-out preference support ahead of the 2027 deadlines.
  • Legislative follow-ons expanding coverage to smaller AI developers.

Sources

[1] "Joint California Policy Working Group on AI Frontier Models" https://www.cafrontieraigov.org/

[2] "The California Report on Frontier AI Policy" https://www.gov.ca.gov/wp-content/uploads/2025/06/June-17-2025-%E2%80%93-The-California-Report-on-Frontier-AI-Policy.pdf

[3] California Legislative Information -- Senate Bill No. 53 "Transparency in Frontier Artificial Intelligence Act" https://leginfo.legislature.ca.gov/faces/billTextClient.xhtml?bill_id=202520260SB53

[4] California Legislative Information -- Assembly Bill No. 1043 "Digital Age Assurance Act" https://leginfo.legislature.ca.gov/faces/billTextClient.xhtml?bill_id=202520260AB1043

[5] California Legislative Information -- Assembly Bill No. 566 "California Opt Me Out Act" https://leginfo.legislature.ca.gov/faces/billTextClient.xhtml?bill_id=202520260AB566

[6] "What is GPC?" https://oag.ca.gov/privacy/ccpa/gpc

[7] GlobalPrivacyControl https://globalprivacycontrol.org/

[8] California Legislative Information -- Assembly Bill No. 56 "Social Media Warning Law" https://leginfo.legislature.ca.gov/faces/billTextClient.xhtml?bill_id=202520260AB56

[9] California Legislative Information -- Senate Bill No. 446 "Data breaches: customer notification" https://leginfo.legislature.ca.gov/faces/billNavClient.xhtml?bill_id=202520260SB446

[10] California Legislative Information -- Senate Bill No. 243 "Companion chatbots." https://leginfo.legislature.ca.gov/faces/billNavClient.xhtml?bill_id=202520260SB243

r3b1s@[redacted]$~/blog/california-policy.mdt hat  ? keys