Safety in artificial intelligence development has become an official topic in Washington D.C., and OpenAI decided to join the conversation in a way few people expected.
In September 2026, Chris Lehane, the company’s global head of corporate affairs at the company behind ChatGPT, took part in a round of discussions with American lawmakers and declared support for a specific part of the FRONTIER Act, a bipartisan bill that requires independent audits of the largest AI models currently in use.
It marks the first time OpenAI has publicly backed a federal requirement of this kind, one that would put a third-party group in charge of evaluating the work done by AI developers.
Previously, the company had supported state-level regulatory efforts but had never put its name behind a mandatory federal oversight proposal.
And that changes the dynamics of the regulatory debate in the country quite a bit. 🧐
What the FRONTIER Act is and why it matters
The FRONTIER Act is a bipartisan bill, meaning it was built with support from lawmakers on both sides of the aisle, which already says a lot about the level of concern in Washington over the rapid advancement of AI models. The proposal is sponsored by Representative Jay Obernolte, a Republican from California, and Representative Lori Trahan, a Democrat from Massachusetts, and it brings together exactly the kind of alliance that gives political weight to an initiative like this.
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
The bill proposes that artificial intelligence systems deemed high-impact undergo independent audits. In practice, an external organization with no direct ties to the developing company would have to determine whether an AI company’s safety protocols are rigorous enough to mitigate what the bill describes as acceptable levels of catastrophic risk. On top of that, the bill would require AI developers to publish reports on their models and disclose security incidents. It is a robust proposal and, for that very reason, heavily debated across the industry.
The logic behind the bill is fairly straightforward: the more powerful an AI model is, the greater the potential for impact, both positive and negative. Systems like ChatGPT and its successors have already demonstrated the ability to influence medical, legal, financial, and even large-scale communication decisions. Without an external verification mechanism, the public has to blindly trust what companies say about the safety of their own products. The FRONTIER Act aims to break that dependency by creating a structured, independent review process that does not rely on the goodwill of any specific organization.
The name of the bill, by the way, is no coincidence. FRONTIER is a direct reference to the idea of the technological frontier — the most advanced systems that exist at any given time. The proposal does not attempt to regulate every single AI tool out there, but rather those operating at the edge of what technology can do today. That is where the risks are greatest and where the need for audits becomes most urgent. For anyone following the industry closely, the bill represents one of the most serious regulatory initiatives that the U.S. Congress has put on the table so far. 📋
Why OpenAI’s support is a game changer
For years, OpenAI navigated the regulatory landscape carefully. The company always championed the idea that artificial intelligence needed to be developed responsibly, but when it came time to turn that rhetoric into concrete support for specific legislation, especially at the federal level, the stance was more cautious. The company’s track record included backing state-level initiatives, which tend to have a more limited reach than a federal law. So when Chris Lehane publicly declared support for a portion of the FRONTIER Act during the meeting with lawmakers, the market, legislators, and competitors all took notice.
An important detail worth highlighting: despite supporting independent audits of AI models, OpenAI did not endorse the bill in its entirety, according to a company spokesperson. In other words, the support is selective and focused specifically on the external verification mechanism, not on every item in the legislative text. That nuance matters for understanding the company’s actual position. 🤝
This move has at least two possible readings, and both are relevant for understanding what is happening in the industry. The first is the more straightforward interpretation: OpenAI genuinely believes that independent audits are necessary to ensure the safety of the most advanced AI models, including its own. This reading makes sense especially considering that the company was founded with a stated mission to develop AI in a safe and beneficial way for humanity. The second reading is more strategic: by supporting structured federal regulation, OpenAI helps establish a standard that could be more favorable to large, established players than to smaller startups. Regardless of the motivation, the practical effect is the same: one of the most influential AI companies in the world is telling the U.S. Congress that federal oversight is welcome.
Lehane’s statements come at a time when researchers and AI developers have been warning about the potential for advanced systems to cause harm to people. On the CBS Mornings show, the executive openly acknowledged the risks of letting increasingly powerful models loose in the world. According to him, the industry has entered a new phase and a new chapter in terms of capabilities, and there is high confidence that it will be possible to build the guardrails around this technology, starting with the companies themselves. 🚀
The United States and leadership in AI safety
Lehane also stated that the United States is well positioned to lead the safety agenda in artificial intelligence and to develop global standards for the use of these platforms. According to him, there is real confidence that the country can rise to this challenge responsibly and in the right way. This kind of messaging reinforces the idea that regulation, far from being a hindrance, can be treated as a competitive advantage when properly structured.
In a blog post from September 9, Lehane had already advocated for regulatory action to govern AI development. OpenAI has been calling on Congress to establish mandatory national safety requirements and argues that leading AI developers should collaborate to create industry standards. According to the executive, the stakes are enormous: advanced AI could accelerate the development of new medications, strengthen critical infrastructure, expand economic opportunities, and help solve scientific problems that have resisted generations of human effort. But he acknowledges that the same capabilities that make models more useful also bring risks, and those risks are not going to stay confined to a few frontier labs.
What independent audits actually mean in practice
Talking about independent audits of AI models might sound abstract, but the process has very real implications for those who develop and those who use these technologies. In practice, an audit like this would involve giving access to external evaluators — which could be specialized firms, research institutes, or accredited government entities — to the system being assessed. These evaluators would run specific tests to identify problematic behaviors, verify whether the model aligns with the guidelines declared by the developer, and document capabilities that could pose risks in certain use cases. The result would be a report made available to the regulator and, depending on the terms of the law, also to the public.
For the most advanced AI models, the auditing process is particularly complex because these systems have emergent behaviors that are not always predictable. A model trained to do one thing can demonstrate capabilities in completely different areas, sometimes surprising even its own creators. Audits need to account for this, testing not only the declared functionalities but also unusual situations, unexpected instruction combinations, and use-case scenarios that were not part of the original development scope. It is highly sophisticated technical work, and part of the debate around the FRONTIER Act revolves around defining who has the expertise to conduct this kind of evaluation reliably and impartially.
From the end user’s perspective, the existence of independent audits represents an extra layer of safety and transparency. Today, when someone uses a service built on advanced AI models, there is no simple way to independently verify whether the developer’s claims about the system’s behavior are accurate. With an established audit regime, there would be an external and verifiable record of the characteristics and limitations of each system. That does not solve every problem, but it creates a much more solid foundation for consumers, businesses, and regulators to make informed decisions about which systems to use and in what contexts to trust them. 🔍
The perspective of those calling for more caution
Not everyone believes companies will act on their own. Joseph Miller, an AI safety researcher and founder of PauseAI UK, a group that advocates for slowing down the development of the technology, argues that the public debate about how to manage the potential risks of AI is long overdue. He also believes that leading developers are unlikely to adopt sufficiently strong measures to safeguard their technologies without government intervention.
According to Miller, the government needs to take the lead on this issue because companies probably would not do it without the public sector stepping in. This perspective contrasts with the optimism of industry executives and shows that there is still plenty of room for discussion about how far self-regulation can actually go.
The debate around the FRONTIER Act and OpenAI’s stance is far from over. Other major players in the industry still need to weigh in, and the details of how audits would be implemented will generate a lot of discussion before any final text gets approved. But the fact that we have reached this point in the conversation is, in itself, a sign that the topic of safety in AI models has moved beyond voluntary promises and has firmly entered the American legislative agenda in a concrete and irreversible way. 💡
