OpenAI has unveiled a new proposal for regulating advanced artificial intelligence systems, creating a notable policy difference with the White House’s recently announced approach to AI oversight.
The debate centers on how advanced AI models should be evaluated, who should oversee those evaluations, and what role the federal government should play in ensuring safety as AI technology continues to develop at a rapid pace.
The new OpenAI AI Safety Rules proposal calls for mandatory evaluations of advanced AI systems before they are widely deployed. However, the company believes these evaluations should be overseen by civilian agencies rather than intelligence organizations.
The proposal differs from a recent executive order issued by President Donald Trump. The White House plan introduced a voluntary framework for evaluating advanced AI systems for cybersecurity risks. Under that approach, the National Security Agency (NSA) would take a leading role in the evaluation process.
OpenAI has instead suggested that the Center for AI Standards and Innovation (CAISI), a division within the National Institute of Standards and Technology under the Commerce Department, should oversee testing and evaluations.
The proposal arrives at an important moment for AI policy in Washington. OpenAI Chief Executive Officer Sam Altman traveled to the nation’s capital for meetings with White House officials and lawmakers from both major political parties.
During discussions with reporters, OpenAI Vice President of Global Affairs Chris Lehane expressed support for the overall direction of the administration’s efforts. He described the executive order as a sign that policymakers recognize the need for a framework to evaluate advanced AI systems.
At the same time, Lehane indicated that OpenAI would encourage lawmakers and administration officials to expand CAISI’s role in AI oversight.
According to Lehane, CAISI has already established working relationships with leading AI companies. He noted that OpenAI, Anthropic, and other major developers currently share information about advanced AI models with the agency.
These existing partnerships, he argued, provide CAISI with valuable experience in evaluating emerging technologies and understanding how advanced AI systems function.
Lehane also raised concerns about another element of the White House plan. The executive order includes a proposal to create a classified benchmarking process for advanced AI models.
Under such a system, government agencies could use confidential standards to determine when a model becomes powerful enough to require additional scrutiny.
OpenAI believes companies need greater clarity regarding these thresholds. Lehane suggested that developers should understand what capabilities trigger evaluations and how those decisions are made.
He said establishing transparent criteria would help companies better prepare for compliance and improve trust in the evaluation process.
The company sees the current period as an opportunity to shape future AI policy. The executive order gives federal agencies 60 days to develop additional details, leaving room for discussions among government officials, industry leaders, and lawmakers.
Lehane noted that several members of Congress are also considering proposals that could strengthen CAISI’s authority and expand its ability to conduct AI evaluations.
One area where OpenAI appears to support stronger oversight is mandatory testing for advanced AI systems. The company believes evaluations should not be left solely to individual developers.
Lehane said no single AI company should have the authority to decide on its own whether a system requires safety testing. Instead, he argued that a structured and standardized evaluation process should apply across the industry.
However, OpenAI stopped short of supporting more aggressive regulations. Company representatives expressed caution about proposals that would require government approval before releasing new AI models.
According to Lehane, policymakers should first establish effective evaluation systems before considering additional restrictions on AI development.
The discussion highlights the growing challenge facing governments around the world. As artificial intelligence becomes more powerful and widely used, regulators must balance innovation with safety, transparency, and public trust.
The OpenAI AI Safety Rules proposal reflects the company’s view that mandatory testing, clear standards, and civilian oversight can help achieve that balance. Meanwhile, federal officials continue to develop their own framework for managing the risks and opportunities associated with advanced AI systems.
With discussions now underway in Washington, the coming months could play a major role in shaping how artificial intelligence is governed in the United States and beyond.
