Jose Felgueroso
Abogado | Attorney

  • Home
  • Blog


California AI Bill Vetoed by Governor Newsom

info@josefelgueroso.com

Versión en español

2024-11-14

Introduction

California Governor Gavin Newsom vetoed California Senate Bill 1047, known as the "Safe and Secure Innovation for Frontier Artificial Intelligence Models Act." The bill sought to establish safeguards for AI model development to mitigate catastrophic risks. Newsom's decision reflects the challenge of regulating an evolving technology sector.

Background and Scope of the Bill

SB 1047 aimed to regulate developers of "covered models," large-scale artificial intelligence systems requiring significant computational resources for training. It proposed the creation of a state-level Board of Frontier Models to oversee AI model development and ensure compliance with safety protocols.

The Bill's Requirements Regarding Testing and Safety

The bill introduced stringent requirements for AI developers, including:

  • Implementation of protocols for an immediate system shutdown in emergencies.
  • Submission of a comprehensive safety and security plan to the state Attorney General.
  • Mandatory third-party audits of compliance with safety measures, with reports accessible to regulators.
  • Regular reporting of any AI safety incidents, including unauthorized access or critical failures.

These provisions targeted the mitigation of risks associated with advanced AI, such as misinformation, deepfakes, and critical infrastructure vulnerabilities.

Comparing Risk-Based Approaches

SB 1047 adopted a risk classification system based on computational capacity, focusing on large-scale models exceeding specified thresholds of processing power (10^26 FLOPS) and training cost ($100 million). However, Governor Newsom's veto message explicitly criticized this approach, noting that smaller, specialized models could be equally or more dangerous than the large-scale systems targeted by the bill. Instead, Newsom advocated for an empirical, evidence-based framework that would evaluate AI systems based on three key criteria: deployment in high-risk environments, involvement in critical decision-making processes, and use of sensitive data.

The alternative use-based approach aligns with emerging regulatory frameworks worldwide. For instance, both Colorado's SB 205 and the European Union's AI Act employ risk-based classifications that focus on use cases and potential impacts rather than computational resources. Such frameworks recognize that AI risks stem primarily from how systems are deployed and configured, not merely their size or processing power. The governor's position emphasizes the need for "proactive guardrails" while maintaining sufficient flexibility for innovation. This balanced approach would allow for more targeted oversight of AI applications in sensitive domains such as law enforcement, healthcare, and critical infrastructure, regardless of the underlying model's size or computational requirements.

Penalties

Non-compliance with the bill's provisions would have exposed developers to civil actions initiated by the Attorney General. The bill also included whistleblower protections to encourage reporting of safety lapses, underscoring its emphasis on accountability within the AI sector.

Conclusion

While acknowledging the importance of regulating artificial intelligence, Governor Newsom emphasized the need for adaptable, evidence-based frameworks. His veto underscores the challenges in balancing innovation with safety and accountability in an era of rapidly advancing AI capabilities. As California continues to deliberate on AI regulation, the dialogue initiated by SB 1047 will undoubtedly shape future policies in this area.

Sources

SB-1047 Safe and Secure Innovation for Frontier Artificial Intelligence Models Act

Governor Newson's Statement about his Veto of SB 1047