A Critique of Transparency Provisions in NY’s RAISE Act (1.0)

2025-07-08 · Source: AI Accountability Review · Field: Legal & Regulatory — Regulatory Affairs & Government Relations, Compliance & Risk Management, AI Regulation · Depth: Intermediate, medium

Summary

New York's "Responsible AI Safety and Education" (RAISE) Act aims to regulate "frontier models" by mandating the disclosure of specific information for oversight. The law defines a frontier model as one trained with over 10^26 computational operations and a compute cost exceeding $100 million, a definition criticized for being arbitrary, subject to change, and relying on undisclosed data from model developers. A core mechanism is the "safety and security protocol," a transparency report developers must create before deployment and update annually. This protocol, shared with administrative bodies like the attorney general and made publicly available in redacted form, details organizational procedures, sociotechnical measures, and testing to mitigate "critical harm" (defined as $1 billion+ in damage or 100+ lives lost). While the act ensures accessibility, relevance, and timeliness of information, it lacks provisions for third-party auditing to verify accuracy, relying instead on developers not knowingly making false statements. The law includes carve-outs for privacy, trade secrets, and academic research.

Key takeaway

For CTOs and VPs of Engineering evaluating AI governance frameworks, New York's RAISE Act highlights the critical need for precise, auditable definitions of regulated AI systems. Your teams should prioritize developing internal transparency protocols that can withstand external scrutiny, even if current legislation has definitional gaps, as future laws will likely strengthen accuracy and understandability requirements. Be prepared for evolving regulatory landscapes that demand verifiable data on model training and safety.

Key insights

AI governance through transparency faces challenges in defining regulated entities and ensuring disclosed information's accuracy and understandability.

Principles

• Transparency requires clear, auditable definitions.
• Accountability needs designated responsible persons.
• Prospective accountability aims to prevent future harm.

Method

Large model developers must create and annually update a "safety and security protocol" detailing mitigation and testing procedures for critical harm, sharing it with administrative bodies and the public.

In practice

• Define "frontier models" with verifiable metrics.
• Implement third-party audits for protocol accuracy.
• Standardize transparency report formats for clarity.

Topics

• AI Regulation
• Frontier Models
• AI Transparency
• AI Safety
• New York RAISE Act

Best for: CTO, VP of Engineering/Data, Director of AI/ML, Policy Maker, AI Ethicist, Legal Professional

Related on AIssential

• New York Signs off on AI Safety Legislation — New York's RAISE Act establishes state-level AI safety regulations, challenging federal preemption efforts.
• Transparency and Accountability Gaps in Trump's New AI Executive Order — Classified AI model evaluation frameworks undermine public trust and democratic accountability, especially when led by intelligence agencies.
• Govt body plans legal framework for AI companies, tech oversight - Business Standard — India is creating a unified legal framework to regulate AI companies and their products within its borders.
• Trump loses more control over AI regulation as Illinois passes landmark law — Illinois's new AI safety law mandates third-party testing and incident reporting for frontier models, setting a national regulatory precedent.
• AI Roundup 150: Empire State Defiance — State-level AI regulations are intensifying, challenging federal efforts and influencing industry standards.
• OpenAI unveils security framework, citing emerging regulations — AI providers are proactively establishing governance frameworks to meet diverse, rapidly evolving state and international regulatory demands.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI Accountability Review.