The leaders of Anthropic, Google, Meta, OpenAI, Nvidia and xAI recently joined President Trump at the White House to sign a voluntary accord on artificial intelligence. Its commitments are concrete: stronger internal controls, independent outside evaluation and board-level review.

The harder question, which remains to be seen, is whether it will earn the confidence of the people expected to live and work with these systems, who have been increasingly resistant to AI proliferation.

AI ISN’T JUST A PROBLEM, IT’S A SOLUTION. SOME PEOPLE ARE ALREADY MAKING THE MOST OF IT I often say that one of the biggest challenges we have to fear about AI is the fear of AI itself. That fear is becoming a national liability.

In a new EqualAI survey conducted by YouGov, 46% of Americans said they would be uncomfortable with a company using AI in a way that involves them or their data, regardless of safeguards. Yet 71% said a positive experience with an AI system would increase their confidence in it, and 63% said the same of a positive evaluation by independent scientists.

That 63% is an important window into mechanisms to trust — and offers a bridge to build it. People do not want assurances; they want verification.

AI will deliver for our economy, national security and competitiveness in the ways we are envisioning only if Americans actually use it. A nurse who does not trust an AI-enabled diagnostic tool will not rely on it. A manufacturer that cannot understand what its AI agents are doing will not deploy them at scale.

AI 'DOOMERS' RISK KNEECAPPING US VS CHINA, BUT EX-PENTAGON VET SAYS BOTH EXTREMES MISS THE MARK That is why we need to be thoughtful when discussing "pause," because if it does not include capable competitors, such as Chinese companies, the technology continues to advance but elsewhere and without the guardrails we’d like to establish. The question we need to answer is: what will make Americans confident enough to adopt AI?

The debate is too often framed as more rules versus fewer rules. A more useful dialogue centers on how we can build the governance capacity to let AI scale.

The public is telling us we have work to do. A recent Reuters/Ipsos poll found that 73% of Americans believe AI companies have not done enough to prevent serious harm from their technology; 55% said they would support slowing AI development. The EqualAI polling points to what could change that equation: 63% want independent scientific review of AI systems, while 60% support third-party audits. Sixty-eight percent say it is essential to be able to correct wrong information, and 65% say the ability to appeal an AI-driven decision is essential.

When I testified before the House Select Committee, I heard strong, bipartisan interest in getting this right. My message was straightforward: American leadership on AI requires leadership in AI governance. Congress can use authorities it already has, including procurement, agency governance, and consumer protection, as well as clarifying AI-specific protections, such as establishing definitions and protocols for AI incident reporting, as they’ve done for cyber incidents, and defined accountability for AI use cases with higher risk levels and making AI literacy a national competitiveness priority.

And companies don't need to wait for Washington. Every leadership team deploying AI should ensure they’ve taken these four steps: First, make AI visible. You cannot govern what you cannot see. Organizations should know the Who, What, Where, When and How: where AI is deployed, what it can do and access, who owns it, when it will be tested and how unexpected outcomes will be addressed.

WHEN AI TESTS CAUSE DAMAGE, WE NEED STRONGER SAFEGUARDS AND REAL ACCOUNTABILITY Second, establish independent evaluation. The institutional model is open to debate. The underlying questions are not: Who evaluates AI systems? Against what benchmarks? And what happens when a system fails?

Third, create a system to identify and handle significant AI incidents. Define incidents and ensure all responsible parties understand escalation pathways and reporting obligations. We also need the federal government to establish these Fourth, require accountability. Every organization should have a point person for deployments in certain risk categories. As AI becomes more agentic, systems can take actions rather than simply generate information. Responsible officers should ensure an agent only has the minimum access required.

Finally, ensure your workforce and families are AI literate. People do not need to become engineers. They need to understand what AI can and cannot do, when to question it and when human judgment must remain in charge.

We have navigated technological inflection points before. Cars did not become less innovative when we established licenses, traffic lights and global safety standards. Those institutions made mass adoption possible. AI will be no different. Governance accelerates trust and as a result, adoption.

The answer is not to slow technological progress until we eliminate every risk. Nor is it to tell Americans to trust AI just because we want them to. As we accelerate AI development and capabilities, we need to match pace with the mechanisms that enable our capacity to govern it.

America's advantage has never been simply building powerful technologies. It is that we build along with the institutions and standards that allow powerful technologies to be trusted and adopted at scale. This moment asks us to do that. again.