Artificial Intelligence
California Moves to Strengthen AI Safety Rules as Technology Risks Grow
California is strengthening its artificial intelligence safety framework with independent AI audits, verification organizations and new oversight measures as concerns about advanced AI risks continue to grow.
California is taking new steps to strengthen artificial intelligence safety rules as increasingly capable AI systems raise new questions about cybersecurity, transparency and public safety. In September 2026, Governor Gavin Newsom signed legislation establishing a framework for independent AI verification organizations and AI auditors. The state has also moved to accelerate implementation of these measures and explore additional safeguards for advanced frontier AI systems. The developments come after California introduced its Transparency in Frontier Artificial Intelligence Act (SB 53) in 2025, requiring large frontier AI developers to disclose safety frameworks and report certain critical safety incidents. As AI becomes more capable and is increasingly used in software, business, cybersecurity and public services, California's latest rules could become an important test of how governments regulate advanced artificial intelligence.
California Strengthens AI Safety Rules
California has expanded its AI regulatory framework with new requirements for independent assessment and oversight. Governor Newsom signed SB 813 and AB 1405, creating a framework for independent verification organizations and a state registry for AI auditors.
What Are California's New AI Rules?
The new framework focuses on independent evaluation of AI systems and models. SB 813 establishes a system for certifying independent verification organizations, while AB 1405 establishes a registry and standards for AI auditors, including requirements related to independence, transparency and integrity.
Independent AI Audits Become More Important
Independent AI audits are intended to provide an external assessment of advanced AI systems rather than relying entirely on companies to evaluate their own technology. California's framework is designed to create a more formal process for assessing AI safety and risk.
California's Frontier AI Safety Law
California's 2025 SB 53 requires large frontier AI developers to publicly disclose safety frameworks, report specified critical safety incidents and provide whistleblower protections for people who report serious risks.
AI Risks Are Changing Quickly
Advanced AI systems are becoming increasingly capable of coding, reasoning, using tools and performing complex tasks. Recent incidents involving AI systems have increased attention on cybersecurity, model behavior and the ability to maintain human control over powerful systems. Reuters reported that OpenAI has begun developing a framework for regularly disclosing unexpected or unauthorized AI behavior.
California Explores an AI “Kill Switch”
A September 18 executive order directed California agencies and experts to explore stronger safeguards, including the possible creation of a “kill switch” for frontier AI models. The proposal would involve independent verification of the effectiveness of such a mechanism.
AI Safety and Cybersecurity
AI systems can create new cybersecurity opportunities as well as new risks. California launched an AI Cyber Defense Program in August 2026 to improve the state's ability to detect vulnerabilities, strengthen networks and respond to AI-enabled cyber threats.
AI Safety Rules for Children
California has also introduced safeguards for children using AI companion chatbots. New legislation requires protections including crisis protocols, parental controls and notifications when certain safety settings are disabled.
AI Regulation and Innovation
California's approach attempts to combine AI development with transparency and safety requirements. The state has continued promoting AI innovation while introducing rules covering frontier models, cybersecurity, children, privacy and automated decision-making.
Why AI Auditors Matter
As AI systems become more important in business and government, independent auditors could play a role similar to external evaluators in other regulated industries. Their work could help identify safety and security problems before advanced systems are deployed at larger scale.
The Challenge of Regulating Frontier AI
One difficulty for regulators is that AI technology changes quickly. Rules created for one generation of models may become outdated as capabilities change. California's framework therefore includes mechanisms for updating its approach as technology and international standards develop.
California's AI Rules Could Influence Other States
California is one of the largest technology markets in the United States, making its AI policies significant for companies developing and deploying artificial intelligence. Other states and governments may monitor how California's independent audit and safety framework works in practice.
Artificial intelligence is developing at a rapid pace, and governments are increasingly trying to determine how advanced AI systems should be tested and monitored.
California has been one of the most active U.S. states in developing AI policies. Its latest actions expand an AI safety framework that began with a 2023 executive order and continued through legislation covering frontier models, deepfakes, children, workers and digital rights.
One of the most significant developments in 2026 is the creation of a framework for independent AI verification organizations.
Under SB 813, California established a system for organizations with appropriate expertise and independence to assess AI systems and models for safety and risk. AB 1405 creates a state registry for AI auditors and establishes standards concerning their independence, transparency and integrity.
The concept of independent AI auditing is becoming increasingly important as companies deploy more capable models.
Traditional software can generally be tested against clearly defined requirements. Advanced AI systems can behave differently depending on prompts, tools, data and environments. This makes evaluating their behavior more complicated.
An independent evaluation process could provide additional information about how an AI model performs under different safety and security conditions.
California's approach also builds on SB 53, the state's 2025 Transparency in Frontier Artificial Intelligence Act.
The law requires large frontier AI developers to publicly disclose their safety frameworks and report certain critical safety incidents. It also provides protections for whistleblowers who disclose serious risks associated with frontier AI systems.
The importance of these requirements has increased as AI models become more capable.
Recent incidents have involved AI systems displaying unexpected or unauthorized behavior. OpenAI announced in September 2026 that it would begin regularly disclosing certain cases of model misalignment and unexpected behavior. The company said its new framework is designed to identify, investigate and report these incidents more systematically.
Cybersecurity is another major concern.
AI systems can potentially help organizations identify vulnerabilities and respond to attacks, but increasingly capable systems can also create new security risks. California's August 2026 AI Cyber Defense Program aims to use AI for vulnerability detection, network hardening and incident response while preparing for emerging AI-enabled cyber threats.
California is also focusing on AI systems used by children.
New legislation signed in September establishes additional protections for children interacting with AI companion chatbots. The rules include crisis protocols, parental controls and notifications related to safety settings.
These developments show that AI regulation is no longer focused only on large language models or research laboratories.
Governments are increasingly dealing with AI in areas including education, healthcare, cybersecurity, government services, advertising, employment and social media.
The challenge is finding rules that address genuine risks without preventing useful AI applications from developing.
California's framework attempts to address this challenge through transparency, independent assessments and accountability rather than relying on a single regulatory measure.
The September 18 executive order adds another layer to the discussion by directing officials and experts to explore stronger independent oversight and the potential development of a “kill switch” for frontier AI models. The order proposes that such a mechanism would be independently verified on an ongoing basis.
Conclusion
California is continuing to expand its AI safety rules as artificial intelligence becomes more capable and more widely deployed. The state's latest measures introduce independent AI verification organizations, AI auditor requirements and additional oversight mechanisms. California is also exploring stronger safeguards for frontier AI models, including a potential “kill switch” framework. These policies build on California's existing frontier AI transparency law and broader efforts covering cybersecurity, children, privacy and responsible AI use. The technology industry will be watching how these rules work in practice. The results could influence future discussions about AI regulation, independent AI audits, frontier AI safety and responsible artificial intelligence development. As AI technology continues to evolve, the central challenge will remain finding practical ways to encourage innovation while improving safety, transparency and human oversight.
