California passes AI safety legislation with support from OpenAI and Anthropic

California Governor Gavin Newsom has signed a series of AI regulatory bills aimed at enhancing safety risk assessments, transparency, and cybersecurity for large-scale models. The legislation, backed by OpenAI and Anthropic, addresses concerns over AI misuse in critical infrastructure and deepfake manipulation. The laws require developers to conduct red team drills, establish risk mitigation protocols, and submit third-party audited security reports before deploying trillion-parameter models. The bills focus on creating transparent pre-risk assessment frameworks and industry benchmarks rather than imposing strict technical caps. OpenAI and Anthropic had previously warned against overly rigid provisions that could stifle innovation. The final version balances stringent security measures with industry collaboration, gaining support from major AI research institutions. The legislation also includes traceability standards for AI-generated content, requiring digital fingerprints on deepfake media. It provides legal protection for employees who report security vulnerabilities, prohibiting harsh confidentiality agreements. This marks California's role as a regulatory pioneer in AI safety, balancing security with innovation.