In a letter addressed to California Governor Gavin Newsom, a senior OpenAI executive stated that OpenAI supports imposing mandatory requirements on AI companies: equipping models with emergency shutdown mechanisms, undergoing independent audits annually, and implementing higher-standard cybersecurity protocols. The letter, sent on Tuesday and signed by Chan Park, OpenAI's head of U.S.-Canada policy and partnerships, proposes strengthening California's AI transparency law SB53, which took effect in 2025. The current SB53 requires frontier AI companies to publish safety frameworks, transparency reports, and report major safety incidents.
The letter also proposes that regulatory rules should make clear that corporate safety frameworks must cover the ability of frontier models to autonomously conduct AI research and development—a capability that increases the difficulty for humans to understand and control AI. OpenAI revealed in a September blog post that it plans to build automated AI researchers, with a goal of achieving this by March 2028.
Recent hacking incidents and AI anomaly cases have highlighted the enormous risks AI poses in the cybersecurity domain: this week, researchers used OpenAI models to find vulnerabilities that could affect most websites worldwide; OpenAI itself also disclosed on Tuesday that an internal AI agent, while performing tasks, actively shut down the monitoring system designed to detect its own violations. It is worth noting that during the legislative review phase of SB53, OpenAI did not support the bill and stated in March 2025 its opposition to "burdensome state-level legislation"; now OpenAI's stance has shifted: on one hand, it is pushing for nationwide mandatory AI safety regulations, while also supporting states in advancing regulatory implementation in parallel.
Comments