Sam Altman, the chief executive of OpenAI, told lawmakers on Capitol Hill that the government should require artificial intelligence companies to submit their models for mandatory safety evaluations before release. The proposal, made during a closed-door hearing Wednesday, puts the company behind one of the stronger forms of AI regulation now under discussion in Washington.
What Altman proposed
Altman argued that a federal agency — likely the National Institute of Standards and Technology or a new AI-specific regulator — should set up a testing regime that all major AI developers would have to follow. Under his vision, companies would not be allowed to deploy a new model until it passes a set of predefined benchmarks for things like bias, security, and potential misuse. Those benchmarks would be updated as the technology evolves.
The OpenAI chief did not offer a detailed timeline or budget for such a program, but he stressed that voluntary commitments are not enough. The industry has already seen rapid advances in large language models and image generators, and Altman believes the risks — from disinformation to cyberattacks — demand binding rules rather than optional guidelines.
Why he argues for government oversight
Altman has spoken publicly before about the need for regulation, but Wednesday’s testimony marked his most explicit push for mandatory checks. He told the panel that while OpenAI runs its own internal evaluations, the process lacks independent verification. Without a government-mandated standard, he said, companies have an incentive to cut corners or to downplay risks they detect.
Lawmakers on both sides of the aisle have been wrestling with how to regulate AI without stifling innovation. Some have expressed concern that heavy-handed mandates could push development overseas. Altman acknowledged that tension but argued that the United States can lead by setting a global standard for safety. He pointed to the Federal Aviation Administration’s role in certifying aircraft as a model: one national authority that everyone must work through.
The CEO also warned that if the U.S. does not act, other countries will fill the vacuum with their own rules — rules that might not align with American values on free speech or privacy. That argument resonated with several members of the committee, according to aides who attended the session.
Reaction from lawmakers
The hearing was not open to cameras, but attendees described a serious, mostly bipartisan tone. Several senators asked technical questions about how a mandatory evaluation system would work in practice: who pays for it, how fast can it scale, and what happens when a model fails. Altman offered few specifics, saying the details should be left to regulators and technical experts.
Some lawmakers remained skeptical. A few raised concerns that mandatory testing could slow down the pace of innovation, especially for smaller startups that lack the resources to navigate a federal approval process. Altman pushed back, noting that security failures at big companies hurt the entire ecosystem and that regulation would ultimately build public trust, which the industry desperately needs.
No legislation was introduced during the hearing, but staffers from the Senate Commerce Committee said they are working on a bipartisan bill that could include mandatory evaluation requirements. The timeline for that bill remains unclear.
What comes next
The White House has already issued an executive order on AI that calls for voluntary safety commitments from major developers. Altman’s testimony goes further, urging Congress to codify those commitments into law. The next step is likely a formal markup of a draft bill, possibly later this year.
For now, the question that hangs over the room is one of enforcement: even with mandatory evaluations, can the government keep up with a technology that changes every few weeks? Altman did not answer that question directly. He said only that the alternative — doing nothing — is no longer an option.
