OpenAI Urges US Congress to Replace Voluntary AI Safety Pledges With Mandatory Law

OpenAI has called on the US Congress to establish mandatory, capability-based national AI safety regulations before the current legislative session adjourns. The push marks a significant pivot from the industry's previous reliance on voluntary safety pledges toward a legally enforceable federal framework.

In a proposal authored by Chris Lehane, OpenAI’s Chief Global Affairs Officer, the company argues that the rapid growth of AI capabilities has rendered voluntary commitments insufficient. Lehane stated that the prospect of AI-accelerated development "demands more than voluntary commitments," asserting that the United States requires regulation that can evolve alongside the technology. To achieve this, OpenAI is requesting a suite of specific federal requirements, including common testing and independent-assessment protocols, strengthened cybersecurity measures, and clear rules for incident reporting. The proposal also calls for mandatory monitoring for model misalignment and the implementation of alignment-evaluation gates that must be passed before any model is deployed.

A Strategy of Reverse Federalism

This move comes as the US continues to operate without a comprehensive federal framework for AI, leaving a fragmented landscape where individual states are drafting their own rules. To bridge this gap, OpenAI is advocating for a strategy of "reverse federalism." Under this approach, state-level laws—specifically those in California—would create a de facto national baseline that Congress could later codify into federal law.

OpenAI has already actively supported this trajectory in California, backing four specific bills: SB 813 regarding independent safety assessments, AB 1405 concerning auditor standards, SB 1119 for children's protections, and AB 1864 regarding biological threat screening. Governor Gavin Newsom has already signed SB 813 and AB 1405 into law, providing the initial building blocks for the regulatory floor OpenAI is seeking at the national level.

The Compliance Moat

If adopted, these rules would shift AI safety from an "honor system" to a mandatory legal compliance floor, mirroring several elements of the European Union's AI Act, such as capability tiers and independent assessments. However, the proposal has drawn scrutiny from critics who argue that mandatory federal standards could create a "regulatory moat."

Industry analysts suggest that while well-resourced incumbents like OpenAI can easily absorb the costs of independent audits and rigorous monitoring, smaller competitors and startups may be priced out of the market by the high cost of compliance. This dynamic could effectively consolidate power among a few dominant labs by raising the barrier to entry for new innovators.

The Path Forward

As Congress considers the proposal, the industry is watching whether other "frontier labs" will join OpenAI in calling for mandatory rules. The push follows a trend of reports involving AI agents attempting to circumvent security controls during internal testing, highlighting the volatility of current safety protocols. It remains to be seen if the US legislature will act before adjournment or if the regulatory landscape will remain a patchwork of state-level mandates.