The window for AI policy is open. Act now

Chris Lehane, OpenAI's Chief Global Affairs Officer, urged policymakers to act immediately in light of the exponential advancement of AI capabilities. He emphasized prioritizing substantive safety measures and capability-based regulation over comprehensive regulation.

OpenAI officially endorsed four bills passed by the California State Legislature: SB 813, AB 1405, SB 1119, and AB 1864. These bills cover the establishment of independent safety evaluation infrastructure, the creation of AI auditor standards, the mandate for youth protection mechanisms, and the prevention of AI-based biological threats. Lehane described such state-level legislation not as a replacement for federal regulation, but as a 'reverse federalism' approach to fill gaps and raise standards before federal standards are established.

Internal Safety Measures and Recursive Self-Improvement

Internally, OpenAI is strengthening safety measures, including reinforcing the isolation of frontier research workloads and expanding model behavior monitoring. In particular, for the Astra model, it has introduced universal monitoring of full trajectories and mandatory alignment evaluations prior to deployment. Chief Scientist Jakub Pachocki warned that extreme caution is required regarding the possibility of recursive self-improvement, stating that fully autonomous RSI has not occurred and that OpenAI will not pursue it until safety is assured.

Regulatory Scope and International Standards

OpenAI argues that federal regulation should apply only to a small number of resource-rich labs developing frontier systems, and proposes excluding startups and open model developers. It also added that voluntary commitments are insufficient, and that common testing and independent evaluations should be mandated through the Frontier Safety Blueprint. Lehane concluded that robust safeguards are essential for the US to maintain sustainable leadership through not only technological superiority but also trust, and that establishing interoperable international standards is urgent.