Vermont’s proposed rules go well beyond what the Legislature authorized, creating new obligations the law never included and restricting how online services share content with users. That raises serious First Amendment problems for minors and adults alike. NetChoice, at the very least, urges the Attorney General to narrow the rules to match the statute before they’re finalized.
NetChoice Comments for the Record on Proposed Rules Implementing §§ 2449f and 2449g of the Vermont Age-Appropriate Design Code Act
September 29, 2026
The Honorable Charity Clark
Office of the Attorney General
109 State Street
Montpelier, Vermont 05609
Re: Comments on Proposed Rules Implementing §§ 2449f and 2449g of the Vermont Age-Appropriate Design Code Act
NetChoice is a national trade association of technology and internet businesses that promotes free expression and free enterprise online. We have engaged with Vermont lawmakers throughout the legislative process that led to Act 63, including submitting testimony and circulating legal memos, and appreciate the opportunity to continue our engagement by commenting on the proposed rules implementing provisions of Vermont’s Age-Appropriate Design Code Act (“AADC” or “Act 63”).
NetChoice and our members share the Legislature’s goal of protecting minors online. But that goal must be pursued through means that respect the constitutional rights of all internet users – including minors themselves – and that stay within the authority the Legislature actually delegated to the Attorney General’s Office. Our position on Act 63 has been clear from the outset: the law contains several constitutional flaws. While the proposed rules cannot solve those flaws, they can—and do—introduce new problems and exacerbate existing ones.
As explained below, many of the rules’ specific defects mirror problems that federal courts have already identified in the California AADC and in analogous age-verification and design-code statutes; others are unique to how the proposed rules construe the narrower delegation Vermont’s Legislature actually enacted. We urge the AGO to revise the proposed rules as set forth below before they are finalized. Absent revision, we believe, independent of the statute itself, significant portions of the rules will not withstand judicial scrutiny to the detriment of covered businesses, minors, and the AGO’s own enforcement priorities.
I. The Proposed Rules Exceed the Authority the Legislature Delegated to the Attorney General
The Legislature’s delegation to the AGO in this rulemaking is narrow and specific. Under § 2449f(b), the AGO may adopt rules that “prohibit[] data processing or design practices of a covered business that, in the opinion of the Attorney General, lead to compulsive use or subvert or impair user autonomy, decisionmaking, or choice.” Under § 2449g(b), the AGO’s age-assurance rulemaking is limited to identifying “commercially reasonable and technically feasible methods” for age determination, describing appeal processes for age designations, and specifying privacy protections for age-assurance data – while prioritizing user privacy and accessibility over the accuracy of any given method. Neither provision authorizes the AGO to freestand a general duty-of-care regime, to create new categories of per se prohibited conduct untethered to compulsive use or autonomy impairment, or to mandate universal age verification. The proposed rules do all three.
A. The General Duty Provision Implements § 2449c, Not the § 2449f(b) Delegation
The proposed rules’ Applicability and General Duty provisions prohibit covered businesses from designing or deploying any practice that creates a “reasonably foreseeable heightened risk of harm” – language that tracks the free-standing duty-of-care provision in § 2449c, not the narrower compulsive-use/autonomy-impairment standard in § 2449f(b), which is the only provision that authorizes this rulemaking. Section 2449c contains no independent rulemaking delegation. By using § 2449f(b) to implement § 2449c’s substantive duty, the proposed rules convert a specific, bounded delegation into a general grant of authority to regulate any design practice the AGO deems risky. That is precisely the kind of agency overreach that separation-of-powers principles and ordinary administrative law forbid: an agency may not use a narrow rulemaking grant to accomplish what the Legislature did not authorize through rulemaking. The rules’ standard also evades the Legislature’s express limit on that duty. Section 2449c(c) provides that “[t]he content of the media viewed by a covered minor shall not establish emotional distress, compulsive use, or discrimination.” Yet for recommendation systems, feeds, and similar features, the only “heightened risk of harm” a design practice can create comes from the content a minor sees, so the rules reintroduce through design the content-based liability the statute forecloses. We recommend narrowing the General Duty provision so that it tracks § 2449f(b)’s actual text – prohibiting only those practices that, in the AGO’s judgment, lead to compulsive use or subvert or impair user autonomy, decision-making, or choice.
B. The Data Protection Impact Assessment Regime Has No Statutory Basis
The statute contains no reference to impact assessments, pre-deployment documentation, or ongoing compliance review. Yet the proposed rules create a full Data Protection Impact Assessment (“DPIA”) regime – an eight-factor pre-deployment assessment, multiple categories of mandatory documentation, periodic re-review, and a rebuttable presumption of non-compliance where documentation is missing. Other states that impose DPIA obligations on covered businesses – including California, Connecticut, and Maryland – did so through legislation, where the legislature made the affirmative policy judgment that such a regime was warranted and defined its scope. Vermont’s Legislature made no comparable judgment in Act 63. The AGO cannot supply through rulemaking a compliance architecture the Legislature considered and declined to enact, or never considered at all. We recommend striking the DPIA regime in its entirety, or, at minimum, confining any assessment obligation to the specific practices § 2449f(b) authorizes the AGO to prohibit.
The Ninth Circuit has already held that California’s DPIA requirement likely violates the First Amendment (NetChoice, LLC v. Bonta, 113 F.4th 1101 (9th Cir. 2024)). The court reasoned that the requirement compelled businesses to make and report subjective judgments about whether their services could harm children, and then to act on those judgments. In effect, it deputized private companies to police speech on the State’s behalf. The proposed rules’ assessment of “heightened risk of harm” invites the same judgments. It also sits uneasily with § 2449c(c), which excludes content from the Act’s duty of care. A DPIA that requires businesses to evaluate whether their services expose minors to harmful material would accomplish through rulemaking what the Legislature expressly declined to do. We recommend striking the DPIA regime in its entirety, or, at minimum, confining any assessment obligation to the specific practices § 2449f(b) authorizes the AGO to prohibit.
C. Rebuttable Presumptions and Elevated Evidentiary Standards Are Not Authorized
The proposed rules also establish rebuttable presumptions of non-compliance – for example, a presumption that a design practice induces compulsive use whenever it optimizes engagement or lacks a contemporaneous user request – that a covered business must rebut with “clear and convincing evidence.” By creating a rebuttable presumption that designs are harmful, the proposed rules exacerbate the First Amendment problems with the law. In essence, the rules pressure and disincentivize the use of effective communication and speech dissemination tools by saddling companies with the burden to prove their choices do not violate the statute. The AGO cannot, whether directly or indirectly, meddle with the speech market. To do so violates the First Amendment, as numerous Supreme Court precedents make abundantly clear (Moody v. NetChoice, 603 U.S. 707, 741-42 (2024) (the government may not interfere in the speech market to achieve its version of “speech nirvana”); Speiser v. Randall, 357 U.S. 513, 525-26 (1958) (the First Amendment bars the government from doing indirectly what it is forbidden from doing directly); Nat’l Rifle Ass’n v. Vullo, 602 U.S. 175 201 (2024) (same)).
Section 2449h, the Act’s enforcement provision, incorporates the standard chapter 63 enforcement mechanism and includes no burden-shifting language. By creating presumptions and specifying evidentiary standards through rulemaking, the AGO is exercising a legislative function the statute does not confer. We recommend removing all rebuttable presumptions and elevated evidentiary standards from the proposed rules, leaving the AGO to establish violations through the ordinary evidentiary standards that already govern chapter 63 enforcement.
D. The Push-Notification Restriction Rewrites the Statute’s Bright-Line Curfew
Section 2449f(a)(5) prohibits sending push notifications to a covered minor only “between 12:00 midnight and 6:00 a.m.” – a specific, bright-line rule that leaves the AGO no apparent room to broaden through rulemaking. The proposed rules instead restrict push notifications based on personal, behavioral, or inferred data at all hours, not just during the statutory curfew window. Because nearly any minor-directed alert can be characterized as data-based, this provision effectively displaces a clear statutory rule with an open-ended, round-the-clock restriction the Legislature did not enact. We recommend confining the push-notification restriction to the midnight-to-6:00 a.m. window Section 2449f(a)(5) actually specifies.
E. The Purpose-Limitation and Data-Minimization Provisions Rewrite Act 63’s Standards
Section 2449f(a)(1) limits the collection, sale, sharing, and retention of a covered minor’s personal data to what is “necessary to provide an online service, product, or feature with which the covered minor is actively and knowingly engaged.” Section 2449f(a)(2) separately limits the use of previously collected data to the purpose for which the data was collected. The proposed rules replace those context-dependent standards with substantially narrower requirements: processing must be “strictly required” for a “specific feature,” directly responsive to a contemporaneous request, and unavailable through any less data-intensive means. The rules also categorically treat recommendations, profiling, model training, analytics, and other ordinary service functions as impermissible or unnecessary regardless of why the data was collected or how the processing supports the service.
The final rules should instead preserve the statute’s two distinct tests. First, data is necessary to provide a service when it is reasonably necessary to provide, maintain, improve, secure, protect, debug, or ensure the safety, integrity, accessibility, reliability, or quality of the online service, product, or feature with which the covered minor is actively and knowingly engaged. Second, previously collected data should be treated as subject to an impermissible secondary use only when it is processed for a purpose materially different from the purpose for which it was collected. Processing consistent with the disclosed collection purpose—including security, fraud and abuse prevention, child safety, content integrity, accessibility, service improvement, and safety-enhancing ranking—should not become an impermissible secondary use merely because the processing involves profiling, recommendations, analytics, or model development.
The final rules should therefore delete the “strictly required,” “specific feature,” contemporaneous-request, and “less data-intensive means” limitations; remove categorical prohibitions based solely on labels such as recommendations, profiling, analytics, or model training; and assess data use and retention based on the purpose for which the data was collected and the service the minor is using.
II. The Personalization and Recommendation-System Provisions Impose Unconstitutional Content-Based Restrictions on Speech
In Moody v. NetChoice, the Supreme Court confirmed that a website’s choices about how to select, arrange, and prioritize content are protected editorial judgments that the government may not override. The proposed rules’ outright ban on “Behavioral Recommendation Systems” targets exactly this protected function, prohibiting the curatorial act itself regardless of any resulting harm – a textbook speech restriction that the rules make no attempt to satisfy (Reed v. Town of Gilbert, 576 U.S. 155, 163 (2015). See also Moody v. NetChoice 603 U.S. 707, 740 (2024)).
The Ninth Circuit has already enjoined analogous provisions. In NetChoice, LLC v. Bonta, 170 F.4th 744 (9th Cir. 2026), the court affirmed the injunction against California’s AADC data-use restrictions and dark-patterns prohibition, holding that their operative standards – “materially detrimental,” “best interests,” and “well-being” – are unconstitutionally vague because they fail to give businesses fair notice of what conduct is prohibited. In a companion case, the court held that California’s restriction on displaying like counts to minors was content-based and likely failed strict scrutiny. Vermont’s rules rely on similarly open-ended, harm-based standards and would be vulnerable to the same infirmities, sweeping in educational platforms, curated playlists, and navigation features the statutory exceptions at § 2449f(a)(4)(A)–(C) were meant to preserve.
Several related provisions compound the problem. The proposed rules’ Personalized Media exceptions cast doubt on whether user-enabled account settings, not just one-time, contemporaneous requests – satisfy § 2449f(a)(4)’s exceptions; the continuous-media-feed ban has no statutory basis and reaches finite, human-curated catalogs along with unbounded algorithmic feeds; and the “inferred preferences” and model-training restrictions prohibit a business from acting on lawfully obtained information about its own users, which Sorrell v. IMS Health, Inc., 564 U.S. 552 (2011), holds requires heightened First Amendment scrutiny. We recommend, at the very least confirming that account-level settings satisfy the statutory exceptions, exempting finite and human-curated feeds, and withdrawing the recommendation-system ban and inferred-preferences and model-training restrictions.
III. The Age Assurance Rules Should Track the Statute’s Privacy-First, Escalating Framework
Section 2449g(b) directs the AGO to prioritize privacy and accessibility over accuracy and to weigh cost, effectiveness, and the value of reusing existing data and interoperable methods. Together these factors reflect a graduated approach: begin with the least intrusive method – self-declaration or existing account signals – and escalate to more intrusive verification only where necessary. The proposed rules’ mandate that businesses “implement proportionate age assurance measures for any . . . service . . . reasonably likely to be accessed by covered minors” risks being read as requiring credential-based verification as a baseline, which functions as a barrier to protected speech for both minors and adults.
This mandate also functions as an unconstitutional prior restraint. By converting § 2449g(a)’s conditional privacy protections – which apply only “[d]uring the process of conducting age assurance” a business has already chosen to undertake – into a universal deployment requirement, the proposed rules condition a user’s ability to speak or receive speech on first submitting to a government-mandated identity-verification process. The Supreme Court recently recognized that submitting to age verification burdens the exercise of First Amendment rights, and that for fully protected speech, burdens are subject to the same scrutiny as bans.[fn: Free Speech Coal., Inc. v. Paxton, 606 U.S. 461, 482–83, 493 n.12 (2025).] The Court applied a lesser standard only because the law at issue was limited to content obscene to minors. The proposed rules have no such limit: they reach generally available services carrying speech fully protected for minors and adults alike. As the Court has long held, “the Government may not reduce the adult population . . . to reading only what is fit for children.”[fn: Reno v. ACLU, 521 U.S. 844, 875 (1997).
Federal courts have widely enjoined age-verification and analogous mandates on this ground. See NetChoice v. Reyes, 748 F.Supp.3d 1105 (D. Utah 2024), appeal filed, No. 24-4100 (10th Cir.); NetChoice v. Murrill, 812 F.Supp.3d 594 (M.D. La. 2025), appeal filed, No. 26-30016 (5th Cir.); NetChoice v. Griffin, 2025 WL 978607 (W.D. Ark. Mar. 31, 2025), appeal filed, No. 25-1889 (8th Cir.); NetChoice v. Carr, 789 F.Supp.3d 1200 (N.D. Ga. 2025), appeal filed, No. 25-12436 (11th Cir.); NetChoice v. Jones, 822 F.Supp.3d 656 (E.D. Va. 2026), appeal filed, No. 26-1252 (4th Cir.); NetChoice v. Hilgers, 2026 WL 1850018 (D. Neb. June 27, 2026), appeal filed, No.26-2478 (8th Cir.); cf. CCIA v. Paxton, 747 F.Supp.3d 1011 (W.D. Tex. 2024), aff’d on other grounds, 2026 WL 2130729 (5th Cir. July 24, 2026); but see NetChoice v. Fitch, 2025 WL 2078435 (5th Cir. July 17, 2025) (staying preliminary injunction); CCIA v. Uthmeier, 2025 WL 3458571 (11th Cir. Nov. 25, 2025) (same).
The rules’ treatment of age-range data is also inconsistent with § 2449g(a)(2), which permits retaining the age-range determination even after other age-assurance data is deleted. The proposed rules nonetheless bar using that retained determination beyond the initial minor/non-minor check and prohibit reuse across services of the same business, forcing needless re-collection for legitimate, disclosed purposes such as age-appropriate content settings. We recommend confirming that a retained age-range determination may be used for disclosed, age-related purposes and reused across services of the same entity.
IV. Several Key Definitions Are Vague, Overbroad, or Untethered to the Statutory Text
“Compulsive use” requires material disruption of major life activities under the statute, but the proposed rules deem any engagement-optimizing design presumptively compulsive, and the related “heightened risk of harm” definition relies on vague examples like “prolonged use beyond intent” that a federal court has already found unconstitutionally vague in the analogous California context (NetChoice, LLC v. Bonta, 170 F.4th 74). “Subvert or impair autonomy” and related “cognitive influence” language are broad enough to capture ordinary interface design such as a prominent button, a color choice, attractive graphics and compelling storylines – without any objective limiting principle, inviting arbitrary enforcement.
“Necessary to provide a service” is defined conjunctively, requiring processing to be both strictly required and responsive to an express request, while “express and unambiguous request” excludes anything inferred from account settings. Together these mean a minor’s deliberate choice to enable a personalization setting does not count as authorization for it. And “monitoring” is defined so broadly – covering clicks, dwell time, and persistent identifiers – that the rules’ continuous, real-time signal requirement would need to display constantly, defeating its purpose. We recommend making “necessary to provide a service” disjunctive, confirming account settings satisfy “express and unambiguous request,” narrowing “monitoring” to real-time location or communications tracking as well as limiting to individual or parent, and striking undefined, duplicative terms such as “behavioral inference.”
V. The Proposed Rules Should Include an Express Safe Harbor for COPPA-Compliant Practices
COPPA expressly preempts state regulation “inconsistent with the treatment of” activities it governs, 15 U.S.C. § 6502(d), and permits parents to consent to a range of data uses through flexible mechanisms, 16 C.F.R. § 312.5. Under the proposed rules, a business with full, verifiable parental consent could still violate the personalization, model-training, and data-minimization provisions for the same processing – inconsistent treatment the Ninth Circuit has held COPPA preempts. See Jones v. Google LLC, 73 F.4th 636, 642 (9th Cir. 2023). We recommend an express safe harbor confirming that COPPA-compliant processing satisfies the corresponding Vermont requirement.
VI. The Enforcement Framework Should Include Basic Due Process Safeguards
Any enforcement scheme touching protected speech should include narrow, objective standards and a path to seek clarity before facing liability. The proposed rules’ presumptions, elevated evidentiary standard, and vague definitions create considerable uncertainty, with no mechanism for a business to obtain the AGO’s interpretation before risking liability. We recommend an interpretive-guidance process and elimination of the presumptions and evidentiary standard discussed in Section I.B.
Conclusion and Summary of Recommendations
The proposed rules are plagued by serious constitutional infirmities that cannot be cured by minor drafting changes: their core personalization and recommendation-system provisions impose content-based restrictions on speech that cannot satisfy strict scrutiny, requiring the AGO to reconsider its basic regulatory approach to content curation. NetChoice nonetheless offers the following revisions, which are necessary but not, standing alone, sufficient to resolve the Section II speech restrictions:
- Narrow the General Duty provision to track § 2449f(b) rather than implementing § 2449c’s duty of care.
- Remove the DPIA regime and all rebuttable presumptions and elevated evidentiary standards.
- Confine push notifications to the midnight-to-6:00 a.m. curfew, and qualify data-retention limits with “for longer than reasonably necessary.”
- Restore the statutory secondary-use exception and strike the per se secondary-use categories for model training, recommendation, and customization.
- Withdraw the Behavioral Recommendation Systems ban, confirm account settings satisfy the Personalized Media exceptions, and exempt finite, human-curated feeds.
- Confirm reuse of retained age-range determinations for disclosed purposes and across services of the same entity.
- Implement age assurance as a privacy-first, escalating framework, not a universal verification mandate.
- Revise “compulsive use,” “heightened risk of harm,” “subvert or impair autonomy,” “monitoring,” and “necessary to provide a service” to track statutory text and recognize user-enabled settings as valid authorization.
- Add an express COPPA safe harbor and a pre-enforcement interpretive-guidance process.
The timeline compounds these problems. With comments due October 2 and an effective date of January 1, businesses will have only weeks after the rules are finalized to redesign recommendation systems, notifications, data practices, and age assurance. That is not enough time to build for compliance, and it will push businesses to withdraw features from Vermont users rather than risk enforcement. We’d ask the Office to provide an implementation period of at least 12 months after final adoption, or an equivalent enforcement grace period.
We appreciate the opportunity to submit these comments and welcome further discussion with the Attorney General’s Office before the rules are finalized.
Sincerely,
Amy Bos
Vice President of Federal Affairs, NetChoice (The views of NetChoice expressed here do not necessarily represent the views of all NetChoice members.)
NetChoice is a trade association that works to protect free expression and promote free enterprise online.