# Utah House: Artificial Intelligence Transparency Amendments Source: https://www.youtube.com/watch?v=P2wZ8kSjV4o Recap page: https://rapidrecap.app/video/P2wZ8kSjV4o Generated: 2026-02-17T23:04:06.479+00:00 --- ## Quick Overview Utah House Bill 286, the Artificial Intelligence Transparency Amendments, establishes a regulatory framework that shifts AI compliance from self-governance to state-mandated requirements, specifically targeting large frontier model developers with stringent auditing, reporting, and pre-deployment safety plan submissions, while also creating a legal distinction between intentional malicious behavior and accidental errors in AI systems. **Key Points:** - Utah House Bill 286 establishes a new regulatory framework for AI, moving from voluntary commitments to state-mandated requirements for large frontier model developers. - The bill mandates that developers of frontier models with at least $10^{10}$ in annual revenue must submit detailed safety plans before deployment, similar to regulations for hazardous material handling. - The bill defines two risk buckets: catastrophic risk (physical harm/property damage) and child safety risk, which trigger different compliance obligations. - Section 8 creates a 'transparency enforcement restricted account' to fund expert witnesses, consultants, and technology needed to enforce the act. - Section 5 introduces a 'transparency paradox' by allowing developers to redact information to protect trade secrets, creating a tension between transparency and protection. - The bill forces a shift in operational tempo from the software one to an infrastructure one, requiring pre-deployment testing and external auditing. - Penalties for violations include civil penalties up to $1 million for a first offense and $3 million for subsequent offenses, with a requirement to reimburse the state for legal fees. ![Screenshot at 0:07: The introductory slide frames the discussion around a foundational document coming out of the Utah legislature concerning AI transparency amendments, setting the stage for the regulatory discussion.](https://ss.rapidrecap.app/screens/P2wZ8kSjV4o/00-00-07.jpg) **Context:** The video discusses Utah House Bill 286, the Artificial Intelligence Transparency Amendments, which represents a significant legislative effort to regulate the development and deployment of large artificial intelligence models, specifically focusing on frontier models exceeding a certain computational threshold, aiming to replace the industry's previous reliance on voluntary commitments with enforceable state-level oversight. ## Detailed Analysis The discussion centers on Utah House Bill 286, the Artificial Intelligence Transparency Amendments, which significantly shifts the regulatory approach for AI development, particularly for large frontier models. The bill moves away from voluntary industry commitments toward mandatory state-level regulation, effective May 6, 2026. The core mechanism is the requirement for large frontier developers—defined as entities with at least $10^{10}$ in annual revenue—to create and submit detailed safety plans to the state before deploying high-compute models. This process is analogized to building codes, requiring submission of blueprints (safety plans) for regulatory approval before construction (deployment). The bill splits risks into two buckets: catastrophic risk (death, serious bodily injury, or property damage exceeding $1 billion) and child safety risk. For catastrophic risk, developers must prove their model is safe, which includes implementing kill switches or mitigation measures if safety thresholds are crossed. The bill also introduces a 'transparency paradox' in Section 5, allowing developers to redact information to protect trade secrets, which complicates oversight. Furthermore, Section 8 establishes a special enforcement account funded by penalties to pay for external technical experts and auditors. Penalties for non-compliance are severe, starting at $1 million for a first violation, with subsequent violations costing $3 million, plus the company must reimburse the state for legal fees. The bill aims to empower an agency to act as an inspector, ensuring developers cannot simply claim ignorance if harm occurs. ### HB 286 Overview - Establishes a mandatory regulatory framework for AI frontier models in Utah - Effective date is May 6, 2026 - Targets large developers based on revenue threshold ### Risk Categorization - Splits risks into catastrophic (physical harm/property damage over $1B) and child safety risks - Catastrophic risk mandates pre-deployment safety plans and kill switches ### Transparency Paradox - Section 5 allows developers to redact information to protect trade secrets, creating tension with disclosure requirements ### Enforcement and Auditing - Establishes an enforcement account using fines to pay for external auditors and technical experts - Requires external validation of safety plans ### Penalties for Non-Compliance - Civil penalties start at $1 million for first offense, $3 million for subsequent offenses - Companies must reimburse state legal fees if they litigate and lose ![Screenshot at 0:07: The introductory slide frames the discussion around a foundational document coming out of the Utah legislature concerning AI transparency amendments, setting the stage for the regulatory discussion.](https://ss.rapidrecap.app/screens/P2wZ8kSjV4o/00-00-07.jpg) ![Screenshot at 0:31: The discussion highlights that the bill moves regulation from self-governance to a statutory framework, drawing an analogy to building codes.](https://ss.rapidrecap.app/screens/P2wZ8kSjV4o/00-00-31.jpg) ![Screenshot at 1:09: The speaker details that large frontier developers must submit safety plans before deploying high-compute models, comparing it to submitting blueprints.](https://ss.rapidrecap.app/screens/P2wZ8kSjV4o/00-01-09.jpg) ![Screenshot at 2:28: The speaker points out that the bill defines a 'frontier model' based on computational power exceeding $2^{26}$ operations.](https://ss.rapidrecap.app/screens/P2wZ8kSjV4o/00-02-28.jpg) ![Screenshot at 4:44: The speaker explains that the definition of 'harm' for catastrophic risk is explicitly tied to death, serious bodily injury, or property damage over $1 billion.](https://ss.rapidrecap.app/screens/P2wZ8kSjV4o/00-04-44.jpg)