The GC Brief — Issue No. 4

A weekly read for general counsel and in-house teams: what moved, and why it matters.

In this issue: California’s Legislature votes to end private web-tracking suits under the state wiretap law, and to make the change retroactive; a second bill on the same desk would require a human to corroborate any AI-driven firing, and would shift the burden of proof to the employer in enforcement actions over discipline; the FTC’s new Amazon case is about advertising disclosure, not monopoly; Delaware rewrites its privacy statute and one of the two new laws is already in force; California’s privacy regulator tells data brokers that an honest mistake is still a violation; the Fifth Circuit splits a single insurance policy into foreign and domestic halves and sends only one half to arbitration; the Eleventh Circuit decides two Appointments Clause cases in three days and comes out opposite ways; New York’s $75 billion climate superfund law is held preempted; and a district court blocks EPA from reclassifying California’s Clean Air Act waivers by holding that the Congressional Review Act’s no-review bar does not decide what counts as a rule. Two dates are close: the class certification motion in the Workday AI hiring case is due September 14, and the Governor must sign or veto both California bills by September 30.

1. California’s Legislature has voted to end private web-tracking suits under the state wiretap law — retroactively.

SB 690 was enrolled August 31 and presented to Governor Newsom at 2 p.m. on September 4.¹ It is not yet law — the Governor has until September 30 to act — and it is a short bill. It amends a single section of the Penal Code — section 637.2, the private right of action running through the California Invasion of Privacy Act — to add a new 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 Section 638.51 is the pen register and trap-and-trace provision, and it is the hook on which the last three years of website-tracking class actions have hung: the theory that a chat widget, session-replay script, or advertising pixel is a device that captures dialing, routing, addressing, or signaling information without a court order. Subdivision (d)(2) is the part that changes pending cases: the amendments “apply retroactively to any pending claim in an action commenced within two years before the operative date of that legislation.” Note the direction of that window — it reaches backward from the operative date, not forward. The bill leaves the rest of CIPA alone. Sections 631, 632, 632.7, and 638.50 are untouched, as is the $5,000-per-violation statutory damages figure in section 637.2(a) for everything the bill does not cover.³ No member of either house voted against it; the Senate concurred in the Assembly amendments 39–0.¹

Why it matters. This is the rare legislative fix that reaches the docket rather than the next fiscal year. For a company defending a pen-register claim over its own website, the relevant question is no longer only whether a pixel is a trap-and-trace device; it is whether the case was commenced inside the retroactivity window and whether the claim is still pending on the operative date. That makes the calendar a merits question. It also makes the retroactivity clause the thing plaintiffs will attack first — a legislature can ordinarily withdraw a statutory cause of action it created, but doing so as to claims already filed invites a vested-rights and separation-of-powers challenge, and the answer will not arrive quickly. The narrowness cuts the other way too: a complaint pleading sections 631 and 632 alongside 638.51 loses one count, not the case, and the plaintiffs’ bar has had months to see this coming and to plead around it. Most relevant to: privacy litigation counsel and any in-house team with pending CIPA exposure.

2. A second bill on the same desk would require a human to corroborate any AI-driven firing.

SB 947, the No Robo Bosses Act, was enrolled September 4 and adds a new part to the Labor Code.⁴ The operative rules are compact. An employer “shall not rely solely on an ADS when making a disciplinary or termination decision.” Where the employer relies primarily on an automated decision system’s output, it “shall direct a human to corroborate the decision using data that was collected or used to produce the ADS output or other relevant corroborating or supporting information” — the statute lists supervisory evaluations, personnel files, work product, peer reviews, and witness interviews. If the output cannot be corroborated, or the reviewer concludes it is inaccurate, incomplete, or misleading, the employer may not use it. A written post-use notice must go to the employee at the time the decision is communicated, stating that an ADS was primarily relied upon, that a human corroborated it, and whom to contact. Enforcement runs through the Labor Commissioner or a public prosecutor, with a $500 civil penalty per violation and attorney’s fees. The definition of ADS is broad — any computational process from machine learning, statistical modeling, data analytics, or artificial intelligence that issues a simplified output used to assist or replace human discretionary decisionmaking and materially impacts natural persons — and it expressly excludes spam filters, firewalls, antivirus, identity and access management tools, calculators, and databases. The part becomes operative July 1, 2027. A collective bargaining agreement can displace it, but only one that explicitly waives the part in clear and unambiguous terms, expressly provides for wages and working conditions, and provides protection from algorithmic management. This is a revised version of a bill the Governor vetoed once already, and unlike SB 690 it did not pass unanimously: the Assembly voted 53–14 and the Senate concurred 28–10.

Why it matters. The provision that will be felt is not the notice requirement but section 1526.1(c), which shifts the burden: once it is shown that an ADS was used to make a disciplinary or deactivation decision, the employer must demonstrate either that it did not primarily rely on the system or that it complied. There is no private right of action — enforcement runs through the Labor Commissioner or a public prosecutor — and the burden-shifting clause reaches discipline and deactivation rather than termination. That converts an evidentiary question the employer would normally control into an affirmative showing it has to be able to make later, from records it has to have kept. The practical effect is that “a human reviewed it” stops being a description of workflow and becomes a documented artifact with a date on it. Two other features are worth marking. The exclusion list is a definitional safe harbor, and it is short — scheduling, productivity scoring, and attendance analytics are not on it. And the carve-out for CCPA-covered businesses subject to the California Privacy Protection Agency’s automated decisionmaking regulations means a large employer may end up sequencing two different compliance regimes against the same tool. Most relevant to: employment counsel and AI governance leads at California employers.

3. The FTC’s new Amazon case is a disclosure case, not an antitrust case.

On August 31 the FTC and 22 states sued Amazon in the Western District of Washington over the pricing of Sponsored Ads.⁵ The claim is that Amazon represented its advertising auction as a second-price auction — the winner pays just above the runner-up’s bid — while applying an undisclosed surcharge, so that Sponsored Products advertisers were charged their own winning bid close to 80 percent of the time; for Sponsored Brands, the complaint alleges Amazon secretly sets the price for 70 percent of ad clicks and charges the bid amount half the time.⁵ The complaint alleges the scheme has likely illegally extracted over $20 billion from Amazon’s advertising customers, and puts the affected population at approximately 1.2 million advertising customers, including over 500,000 small and medium-sized businesses.⁶ The counts are Section 5(a) of the FTC Act and parallel state consumer protection statutes. Amazon says the suit is misguided and that it strongly disagrees, and points to a different figure: that advertisers saved over $8 billion from 2021 to 2025 because Amazon prioritized ad relevance over bid price alone.⁷ This is a separate matter from the 2023 monopolization case against Amazon, which is set for trial in 2027.

Why it matters. The interesting thing about this case is what it is not. There is no market definition to fight over and no need to prove monopoly power — the theory is that the company described its own pricing mechanism inaccurately, which is a deception case, and deception cases are shorter, cheaper, and harder to win on a motion to dismiss. Any business that sells through an auction, a dynamic pricing engine, or a bidding interface has now been shown the shape of the exposure: the representation about how the mechanism works is itself a marketing claim, and the gap between the described mechanism and the deployed one is the violation. That gap is usually not created by anyone in the legal department. It appears when an optimization is added to a live system and the help-center copy describing the system is not revisited. Both sides’ dollar figures illustrate the same point from opposite ends — each describes the auction’s effect on advertisers, and neither is the disclosure. Most relevant to: general counsel at platform, marketplace, and adtech businesses.

4. Delaware rewrote its privacy statute — and the second of the two new laws is already in force.

Governor Meyer signed HB 380 and HB 381 on September 2.⁸ HB 380 amends the Delaware Personal Data Privacy Act in several ways, three of which change who is covered and what is protected. It lowers the applicability threshold from 35,000 consumers to 10,000. It expands the definition of sensitive data to include neural data — “data that is generated by measuring the activity of an individual’s central nervous system” — along with financial account numbers, log-in credentials, and card numbers, and the Governor’s office describes additional categories including citizenship and immigration status and pregnancy status. And it narrows the employee-data exemption so that it no longer covers personal data processed in connection with profiling and reports. It also narrows the entity-level financial exemption and adds third-party contracting and diligence obligations. HB 380 is effective January 1, 2027. HB 381 is the shorter bill: it amends the computer security breach chapter to clarify when a business must notify the Attorney General of a breach. It took effect on signing.⁹

Why it matters. The split effective dates are the operative fact and they are easy to miss, because the two bills were signed the same day and announced together. The breach-notification change is law now; the privacy amendments are a 2027 problem. Two of the substantive changes deserve separate attention from the threshold change everyone will notice first. Carving profiling out of the employee-data exemption means HR analytics and algorithmic performance tools lose an exemption they have been relying on in Delaware — a state where a very large number of companies are organized, which is not the same as being in scope, but is close enough that the question gets asked. And neural data is now a defined sensitive category in a state privacy statute, which is a marker of where these definitions are heading rather than an immediate compliance burden for most businesses. Most relevant to: privacy counsel and incident response leads.

5. California’s privacy regulator told data brokers that an honest mistake is still a violation.

On September 3 the California Privacy Protection Agency’s Enforcement Division issued Enforcement Advisory No. 2026-01, on the accuracy of data broker registration information.¹⁰ Registered data brokers must disclose, in their annual registration, the categories of personal information they collect — the advisory’s list begins with personal information of minors and runs through names, dates of birth, ZIP codes, email addresses and phone numbers, citizenship and immigration status, union membership, sexual orientation, gender identity and expression, biometric data, precise geolocation, and reproductive health data — and the categories of recipients with whom they share data, including government and law enforcement, foreign actors, and developers of generative AI. The advisory grounds this in Civil Code section 1798.99.82 and 11 CCR section 7603, which requires “only true and correct responses when submitting the registration information,” and it works through four hypotheticals. The statute it invokes carries an administrative fine of $200 for each day a data broker fails to register as required, which the advisory applies to registrations that are inaccurate rather than merely absent. The sharpest sentence is the one about intent: “The Delete Act does not distinguish unintentional mistakes from intentional misrepresentation: both result in incorrect information.”

Why it matters. A strict-liability disclosure obligation with a per-day fine is a different compliance object from a reasonableness standard, because there is no diligence defense to point at. The advisory is also doing something beyond data brokers. By naming generative AI developers as a recipient category that must be disclosed, the agency has made the registry into a public record of which brokers are selling into model training — which is useful to plaintiffs, to journalists, and to the agency itself in deciding where to look next. The recipient list is a reasonable preview of the disclosure categories a regulator now treats as material, whether or not a given company is a registered broker. The registration form has quietly become a sworn statement about a data supply chain that, at many companies, no single person can currently describe. Most relevant to: privacy counsel, and any business that sells or licenses personal information.

6. The Fifth Circuit split one insurance policy into foreign and domestic halves — and sent only one half to arbitration.

In Transportation Consultants, Inc. v. Certain Underwriters at Lloyd’s, London, decided September 3, the Fifth Circuit addressed a surplus lines property policy subscribed by eleven insurers, some foreign and some domestic, in a Hurricane Ida coverage dispute now five years old.¹¹ Louisiana Revised Statutes section 22:868 prohibits arbitration clauses in insurance contracts. The New York Convention supersedes that prohibition — but only where the Convention’s prerequisites are met, and the fourth of them requires a party who is not an American citizen. The policy’s Contract Allocation Endorsement was decisive: it created separate bilateral contracts between the insured and each subscribing insurer. So the Convention reached the contracts with the foreign insurers, and arbitration was compelled as to them; it did not reach the contracts with the domestic insurers, and as to those, Judge Higginson wrote, “Louisiana law governs the policy’s contracts between Transportation Consultants and the domestic insurers and prohibits enforcement of the arbitration agreements in those contracts.” The domestic insurers’ equitable estoppel argument — that they should ride along on the foreign insurers’ clause — failed: “The Insurers cannot use equitable estoppel to avoid that positive law.” The court did vacate the district court’s order lifting the stay as to the domestic insurers and remanded for reconsideration, so the carriers did not leave empty-handed. The result follows Town of Vinton v. Indian Harbor Insurance Co., decided last year on materially identical policy language.¹²

Why it matters. A single sheet of paper in the policy — an allocation endorsement drafted for reasons that had nothing to do with dispute resolution — determined that the same loss will be litigated twice, in two forums, under two procedural regimes, with the attendant risk of inconsistent outcomes on identical policy language. That is a drafting consequence, not a doctrinal one, and it is visible at placement rather than at claim. For policyholders the case is a reminder that the composition of a subscription panel is a forum-selection decision. For carriers it marks the limit of the estoppel workaround in the Fifth Circuit: a domestic insurer on a mixed panel cannot borrow a foreign co-subscriber’s access to the Convention. The line of Louisiana cases running out of Hurricane Ida has now produced a rule stable enough to price. Most relevant to: insurance coverage counsel and risk managers placing surplus lines property programs.

7. The Eleventh Circuit decided two Appointments Clause cases in three days and came out opposite ways.

On September 1, in United States ex rel. Zafirov v. Florida Medical Associates, LLC, the court held that False Claims Act relators are not Officers of the United States and that the qui tam provisions do not violate the Appointments Clause.¹³ It resolved the case on the second prong of Lucia v. SEC: “relators do not hold a continuing position, which is sufficient to determine that relators are not officers who must be presidentially appointed.” The panel weighed tenure, duration, emolument, and duties, and distinguished Morrison v. Olson on the ground that a relator cannot be replaced mid-case by a successor. It reversed the only district court decision to have struck the provisions down and joined the Fifth, Sixth, Ninth, and Tenth Circuits rather than splitting from them — which lowers the odds of Supreme Court review. It also vacated and remanded for the district court to address the defendants’ remaining Article II arguments, so the Take Care and Vesting Clause questions are still live. Two days later, in Russo v. Secretary, U.S. Department of Commerce, the same court held that at least six of the seventeen voting members of the Gulf of Mexico Fishery Management Council were improperly appointed, because they occupy continuing positions established by law and exercise significant federal authority — including the power to veto the Commerce Secretary.¹⁴ It nonetheless declined to vacate the gag grouper catch-limit rule, which had not been adopted through the veto power, and vacated the district court’s judgment with instructions to enter judgment for the government. For scale on the first case: the Justice Department reported that of more than $6.8 billion in False Claims Act settlements and judgments in fiscal year 2025, $5.3 billion came from qui tam suits.¹⁵

Why it matters. Read together the two decisions describe the actual test, which is not about titles or influence but about whether the position continues after the particular matter ends. A relator prosecutes one case and goes away; a council member sits, votes, and can override a cabinet secretary indefinitely. That distinction is portable, and it is the one to apply to advisory boards, self-regulatory bodies, and standard-setting panels whose members are appointed by someone other than the President — a structure common in industries that were organized around delegated authority long before anyone was auditing it. Russo also shows the remedy question separating from the merits question: the court found the appointments unconstitutional and left the challenged rule standing, because the defect had not infected the particular action. A company hoping to unwind a rule on structural grounds needs the defect to have touched the decision it wants undone. And for defendants in FCA cases, the constitutional argument is now narrower but not gone — it moved from the Appointments Clause to Article II’s Take Care and Vesting Clauses, and it is back before the district court. Most relevant to: government contracts and healthcare compliance counsel, and companies regulated by appointed boards.

8. New York’s $75 billion climate superfund law is preempted.

On August 31 Chief Judge Brenda Sannes granted summary judgment against New York in West Virginia v. James, holding the Climate Change Superfund Act preempted on federal common law and Clean Air Act grounds.¹⁶ The Act, adopted in 2024, established a total cost recovery amount of $75 billion, assessed against entities responsible for more than one billion metric tons of covered greenhouse gas emissions between January 1, 2000 and December 31, 2018, payable in full or in twenty-four annual installments.¹⁷ The analytical spine of the decision is the Second Circuit’s City of New York v. Chevron Corp., which the court read as establishing that “[f]or over a century, a mostly unbroken string of cases has applied federal law to disputes involving interstate air or water pollution.”¹⁸ The court also recounted Chevron’s holding that foreign policy concerns foreclose recognizing a federal common law cause of action reaching emissions originating outside the United States. The plaintiffs were twenty-two states together with industry: the U.S. Chamber of Commerce, the American Petroleum Institute, the National Mining Association, the Business Council of New York State, three coal and gas associations, and a coal company. New York and Vermont both enacted climate superfund statutes in 2024, and both were sued promptly.

Why it matters. The reasoning is broader than the statute it struck down. What defeated the Act was not a drafting choice New York could correct but the subject matter itself — interstate and international emissions sit in a field federal law has occupied for a century, and a state cannot assess retrospective liability there whatever label it puts on the assessment. That logic reaches the pending bills in other states and, by extension, the contingent liabilities that energy companies and their insurers have been carrying against them. The decision is a district court ruling and an appeal to the Second Circuit is the obvious next step, so nothing is settled; what has changed is the direction of the risk and the negotiating posture of everyone pricing it. Most relevant to: energy counsel, and general counsel tracking state climate liability exposure.

9. A court blocked EPA from reclassifying California’s waivers — by holding that the Congressional Review Act’s no-review bar does not decide what counts as a rule.

On September 2 Judge Beryl Howell granted California a preliminary injunction in California v. EPA and denied the motions to dismiss filed by EPA and by intervenor-defendants.¹⁹ On June 12 EPA had announced that it was reclassifying four Clean Air Act preemption waiver orders — the 2009 greenhouse gas standards waiver, the 2013 Advanced Clean Cars I waiver, the 2022 reinstatement of that waiver, and the 2025 small off-road engine waiver — as “rules,” and transmitting them to Congress for review under the Congressional Review Act. The CRA provides that determinations under it are not subject to judicial review. The court held that provision does not reach the antecedent question: “an agency determination as to whether an action qualifies as a ‘rule’ is subject to judicial review under the APA and outside Section 805’s scope.” It found the reclassification final agency action, found California likely to succeed, and ordered EPA to restore the status quo and refrain from reclassifying other waivers.²⁰

Why it matters. The holding is about a procedural move, and the move is the point. Reclassifying an existing agency action as a rule restarts a legislative clock that had already run, and it does so without notice and comment and, on the agency’s theory, without any court being able to look at it. This decision says a court can look at the reclassification even if it cannot look at what Congress then does with it — which converts a one-step maneuver into a reviewable agency action with an administrative record behind it. For regulated businesses the practical consequence is that a long-settled approval, permit, or waiver is not as durable as its age suggests, and that the mechanism for unsettling it may arrive as a recharacterization rather than as a repeal. This is preliminary relief on a likelihood finding, and the reclassification question will be back. Most relevant to: regulatory counsel, and companies whose operations rest on long-standing agency approvals.

Key dates

September 14, 2026 — Rule 23 class certification motion due in Mobley v. Workday, the AI hiring case covered in Issue No. 3. A July 13 order reset the schedule after Workday produced three bias evaluation reports on the eve of certification briefing — the same reports behind the privilege ruling. Experts disclosed with the motion must be deposed by October 5; opposition and Workday’s decertification motion are due November 10; reply December 8; and the hearing is set for March 9, 2027.²¹

September 30, 2026 — Last day for the Governor to sign or veto SB 690 and SB 947 (items 1 and 2). Both passed before September 1 and were presented on September 4, which places them under article IV, section 10(b)(2) of the California Constitution.²² A bill neither signed nor vetoed by that date becomes law without signature.

January 1, 2027 — Delaware HB 380 takes effect, lowering the Personal Data Privacy Act threshold to 10,000 consumers and narrowing the employee-data exemption (item 4). HB 381’s breach-notification changes are already in force as of September 2.

July 1, 2027 — SB 947 would become operative if signed (item 2), including the human corroboration requirement, the post-use notice, and the burden-shifting provision.

2027 — Trial in the FTC’s separate monopolization case against Amazon, No. 2:23-cv-01495 (W.D. Wash.), which is a distinct matter from the advertising case in item 3.

Pending, no date set — On remand in Zafirov (item 7), the district court will take up the Take Care Clause and Vesting Clause arguments the Eleventh Circuit did not reach.