OpenAI has unveiled a new framework for regulating advanced artificial intelligence systems that differs from a recently issued White House executive order, highlighting a growing policy debate in Washington over how the most powerful AI models should be evaluated and monitored.
The proposal centers on mandatory testing of advanced AI systems for potential risks before deployment. OpenAI said the evaluations should be overseen by the Center for AI Standards and Innovation (CAISI), a division of the National Institute of Standards and Technology within the Commerce Department, according to a policy paper discussed by company executives and reported by Politico.
The position differs from President Donald Trump's executive order signed on June 2, which established a voluntary framework allowing developers of advanced AI systems to provide the federal government with early access to their models for cybersecurity reviews. The White House order also places a significant role in the process with the National Security Agency and other national security agencies, according to the executive order and accompanying White House fact sheet.
OpenAI's proposal was released as Chief Executive Sam Altman traveled to Washington for meetings with White House officials and lawmakers from both parties. Chris Lehane, OpenAI's vice president of global affairs, said the company generally supports the direction of the administration's approach but believes CAISI is better positioned to conduct advanced model testing because it already works closely with leading AI developers, Politico reported.
"You also have CAISI that has developed into something and has the capability to really do the type of sophisticated testing," Lehane said during a briefing with reporters.
Policymakers are increasingly focusing on the potential national security implications of advanced AI systems. The White House executive order directs federal agencies to strengthen cybersecurity protections, establish an AI cybersecurity clearinghouse and develop a classified benchmarking process to assess advanced AI cyber capabilities. The administration said the framework is intended to protect critical infrastructure and federal systems while maintaining U.S. leadership in the technology sector.
National security concerns surrounding advanced AI systems have intensified as governments examine how the technology could be used in cyber operations, intelligence gathering and attacks on critical infrastructure. The Trump administration's order specifically cites cybersecurity threats and the need to protect government networks, utilities, financial institutions and other key sectors from malicious actors, according to Reuters.
OpenAI's proposal also differs from the administration's approach on the question of mandatory oversight. While the White House order explicitly states that it does not create any government licensing, permitting or pre-clearance requirement for AI models, OpenAI is advocating for compulsory evaluations of advanced systems. Lehane said the company does not believe individual firms should be solely responsible for deciding whether a model requires review.
At the same time, OpenAI stopped short of supporting broader requirements that would require companies to obtain government approval before releasing new models. Company officials said establishing a robust evaluation process remains the immediate priority.
CAISI has emerged as one of the federal government's main centers for AI testing and standards development. The agency works with private-sector developers on evaluations related to cybersecurity, biosecurity and other national security risks, according to information published by NIST. The center also coordinates with federal agencies including the Department of Defense, Department of Homeland Security and the intelligence community.
