On September 8, this site mapped the six privacy bills Newsom received after the August 31 close of the California session, and the operating advice was: desk bills do not change compliance math — wait for the September 30 deadline, then act on chaptered text. Three weeks later the desk is clear and the waiting is over. What follows is the scoreboard, verified against the chaptered bill texts on leginfo and the Governor's office records, plus the one factual correction this process forced.
The Scoreboard¶
- SB 690 (CIPA web-and-app tracking claims) — signed September 30, 2026; Chapter 976. Operative January 1, 2027, with a retroactive reach into pending litigation. The centerpiece for anyone who runs a public website.
- SB 1000 (California AI Transparency Act rewrite) — signed September 30, 2026; Chapter 861. Urgency statute: in force the day it signed. This is the correction detail — more below.
- SB 923 (deletion-rights expansion) — signed September 27, 2026; effective January 1, 2027.
- AB 883 (Delete Act machinery) — signed September 27, 2026; 30-day cycle begins January 1, 2027.
- AB 1542 (prohibit sale/sharing of sensitive personal information) — VETOED September 27, 2026; the existing opt-out-and-limit model stands.
- SB 1050 (synthetic-performer ad disclosure) — signed September 16, 2026; Chapter 246. It left the desk two weeks before the post-deadline date everyone was watching.
- AB 2561 (privacy settings may not be silently reverted) — signed September 27, 2026. Not in the batch six firm summaries tracked; worth the addition. An operating system or application may not undo a privacy setting a user affirmatively configured, absent narrow exceptions such as a legal order.
The signing list for the September 27 batch is on the Governor's legislative update, which also carried the AB 1542 veto message. SB 690 and SB 1000 were acted on September 30 and are verified from their chaptered texts. SB 923's signature is CalPrivacy's announcement — the agency sponsored it.
SB 690: What Signed, What We Described¶
The chaptered text amends a single section of the Penal Code — § 637.2, the civil-action provision attached to the whole communications-privacy chapter — by adding one subdivision:
(d)(1) An action against a private actor for a violation of Section 638.51 alleged to arise from conduct occurring on an internet website, online application, or mobile application may be brought under this section only by the Attorney General.
(2) The amendments to this section by Senate Bill No. 690 of the 2025–26 Regular Session apply retroactively to any pending claim in an action commenced within two years before the operative date of that legislation.
Read mechanically: after January 1, 2027, a private plaintiff cannot bring a pen-register claim (§ 638.51) over a website or app; the Attorney General holds that lane alone. And subdivision (d)(2) is the "mechanism for dying" the September 8 post predicted for the pixel-lawsuit backlog — it is not a prediction anymore. Suits filed at any point in 2025 or 2026 sit inside the retroactive window, and defendants facing pending claims, including arbitrations, now have grounds to move for dismissal once the operative date arrives.
What did not change, now confirmed in signed law: § 631 (interception) and § 632 (recording) keep their private rights of action — the session-replay and chat-recording liability theories live on untouched. The Attorney General can still enforce § 638.51 against trackers. The CCPA is untouched by this bill.
One precision fix on our own post: the September 8 write-up located the structural change in § 638.51 itself, from the enrolled text. The chaptered statute does the work at § 637.2(d) and reaches § 638.51 violations from there. The effect is exactly what that post described — AG-only enforcement for web/app pen-register claims, retroactive — but the address was wrong, and in law the address matters if you are the one filing the motion.
SB 1000: The Part We Got Wrong¶
The September 8 post stated: signed bills take effect January 1, 2027 "unless they carry urgency statutes; none of the six do." That was wrong for SB 1000. The chaptered text — Chapter 861 — declares urgency with a two-thirds vote and states it "shall go into immediate effect." The Governor approved and it was filed with the Secretary of State on September 30. The rewritten AI Transparency Act has been in force since September 30. Anyone who planned to prepare for a January 1 landing has already missed it.
What the rewrite does, section by section from the chaptered text:
- The one-million-user floor is deleted. A "covered provider" is now "a person that creates, codes, or otherwise produces a generative artificial intelligence system that is publicly accessible within the geographic boundaries of the state" — § 22757.1(e), no threshold. The size carve-out that kept small and community-run providers outside this law is gone.
- The manifest option is deleted. The prior law let users opt into a visible disclosure. That user-side option is gone; the disclosure obligation is latent-only (§ 22757.3(a)).
- The latent disclosure says whether the system created or altered the content — and must carry the covered provider's name, the system's name and version, a timestamp, and a unique identifier; it must be permanent or extraordinarily difficult to remove, compatible with the provider's verification tool, and interoperable with recognized standards (§ 22757.3(a)).
- "AI detection tool" becomes "disclosure verification tool" (§ 22757.2): free to the user, able to assess whether image, video, or audio content (beyond minor modifications) was created or altered by that provider's system, outputting system provenance data and suppressing personal information unless the embedded user consented to it. It must accept URL inputs, be publicly accessible, support programmatic invocation, and the provider may satisfy the whole requirement by pointing users at a compliant third-party tool (§ 22757.2(e)).
- The assistive-technology carve-out has an anti-evasion spike. A GenAI system designed to primarily function as assistive technology is exempt until January 1, 2029 (§ 22757.5(b)) — but falsely representing a system as assistive technology is punished at $50,000 per violation under § 22757.4.1, a section that itself repeals on January 1, 2029. The carve-out is a doorway with a camera over it.
- Licensing carries a 72-hour machinery. A covered provider licensing its system must notify licensees of their obligations; if the provider learns an identifiable licensee modified the system out of compliance, it must either terminate the license within 72 hours or notify the licensee and start an escalation that can end with a report to the Attorney General (§ 22757.3(b)).
- Transitional retroactivity, again. A civil action filed before the rewrite took effect is not maintained if the alleged conduct does not violate the rewritten chapter (§ 22757.4(c)). Claims built on the deleted manifest option or the old detection-tool language during the August 2 – September 30 window of the original Act do not survive if the conduct is lawful under the new text.
For the audience here — people who run infrastructure for communities — point one is the real event. Under the prior act, an organization running its own image-generation service for a community with California visitors needed a million users before the statute noticed it. Now the definition turns on whether a generative AI system is publicly accessible in the state, not on audience size. The text also keeps two buckets separate: "covered provider" and "GenAI hosting platform" (§ 22757.1(h), the entity that makes weights or source available for download). Which bucket a self-hosted install serving generated images falls into — host or provider — is an interpretive question this site will not settle tonight, and neither the courts nor the CPPA have. What is settled: text-only deployments sit outside the operative obligations (the chapter's disclosure requirements attach to image, video, and audio content), and image-generating services open to the public in California now carry the latent-metadata and verification-tool machinery with no grace period, because urgency statutes do not offer one.
The Rest, Brief¶
- SB 923 (signed September 27; effective January 1, 2027): the CCPA deletion right now reaches personal information collected "from or about" a consumer — the broker-sourced, enriched, and inferred data deletion requests used to miss. Online-only businesses must offer a web form or portal alongside an email address for privacy requests. Deletion notices must travel to service providers and third parties that received the data. Suppression lists are the permitted memory of a deletion.
- AB 883 (signed September 27; 30-day cycle from January 1, 2027): data brokers process DROP deletion requests every 30 days instead of 45, and the Attorney General, county counsel, or a city attorney can bring civil actions on behalf of elected officials and judges — that machinery operative July 1, 2027.
- AB 2561 (signed September 27): a user's configured privacy setting must stay configured. The rule targets the specific pattern of an app or OS quietly reverting a protection the user switched on — the pattern the industry built as policy.
- AB 1542 (vetoed September 27): the blanket prohibition on selling or sharing sensitive personal information is dead. California keeps the opt-out-plus-limit model, per the veto message. The sensitive-data category still exists and still triggers CCPA risk assessments; what changes is who can be stopped, which is no one by default.
- SB 1050 (signed September 16; Chapter 246): video and audio advertisements that prominently feature AI-generated performers require explicit disclosure, with platform takedown obligations after court findings.
The Calendar Now¶
- January 1, 2027 — SB 690 operative (retroactive dismissal window opens for web pen-register claims); SB 923 deletion expansion; AB 883's 30-day cycle; the already-scheduled trio arrives: browser opt-out signals under the Opt Me Out Act (AB 566), automated-decision technology notices and opt-outs, and the first cybersecurity audit period for the largest businesses. CalPrivacy's inflation adjustment (thresholds, fines, damages) is also due in January 2027 — odd-numbered years are its cycle.
- July 1, 2027 — AB 883's provisions for elected officials and judges.
- January 1, 2029 — the assistive-technology carve-out ends; the assistive flag joins the latent disclosure; § 22757.4.1's $50,000 penalty repeals.
What This Changes for the Self-Hosted Posture¶
Nothing about the posture — and the reasons are the useful part.
The liability layer narrowed: the private pen-register channel over websites dies January 1 and reaches back to 2025 filings. If your community's pages carry pixels because the demand letters were the only realistic risk, the worst case just got cheaper. But the scripts do not know that. They still do whatever their controllers want, the CCPA still applies, recording theories against session replay and chat widgets still have their private lane, and the Attorney General can still come. A narrowed channel is not a permission.
The data layer tightened: from January 1, a deletion request follows personal information to wherever it came from. Stacks that never ingest third-party-sourced data — typical for self-hosted services — skip this obligation entirely, because the data the statute now reaches is data they do not hold. Organizations quietly running enrichment or broker-fed lists inherit a workflow: intake, search, notify, delete, suppress.
The provenance layer went live early on the one urgency item: public image, video, or audio GenAI in California now carries latent provenance metadata and a free verification path at any audience size. A text-only deployment is untouched today. An image-generating one has homework with no grace period.
September 8's standing line — you cannot audit what you do not run — holds through every layer that moved. What the resolution of this batch adds is the observation underneath it: the liability layer is a legislative calendar, not a property of the internet. Sacramento moved three layers this month; the page that loads no third-party scripts, holds no third-party-sourced personal data, and generates no public synthetic images feels all three moves as nothing.
Correction¶
September 8, 2026, this site wrote that none of the six desk bills carried urgency statutes. Wrong for one: SB 1000 is an urgency statute (Chapter 861, § 8) and took effect September 30, the day it was signed and filed. The reworked AI Transparency Act obligations are live now — the one-million-user floor is already gone. The same post located SB 690's structural change in § 638.51; the chaptered statute makes the change at § 637.2(d). The effects were described correctly in both cases; the dates and addresses in this post supersede the September 8 ones.