PARIS—US President Donald Trump’s decision to put Treasury Secretary Scott Bessent in charge of implementing his recent executive order on AI oversight marks a significant shift in America’s approach to tech governance. Such a change was unlikely to come from the AI industry or sectoral regulators, both of which remain wedded to the familiar narrative that regulation stifles innovation and that the overriding policy priority is to beat China by developing the world’s most advanced AI models—including, ultimately, superintelligence.

 

While Bessent may share the administration’s determination to defeat China, placing the Treasury Department at the center of AI oversight in the United States brings a welcome institutional perspective. Seen through the lenses of national security, cybersecurity, financial stability, and critical infrastructure, AI calls for a very different regulatory philosophy from the hands-off, free-market approach that has defined Trump’s AI policy until now.

 

Bessent’s latest proposal is the clearest sign of that shift. According to Bloomberg, the Trump administration could soon create an agency modeled on the Financial Industry Regulatory Authority (FINRA), an industry-funded self-regulatory organization that supervises broker-dealers under a framework authorized by Congress and overseen by the Securities and Exchange Commission.

 

The good news is that, for the first time, a US administration is seriously considering a dedicated body to protect the public interest by evaluating frontier AI models before they are released. The problem is that the proposed agency is modeled on FINRA, which is governed and funded by the very industry it regulates. It is hardly surprising, then, that Google DeepMind CEO Demis Hassabis has long advocated this approach.

 

Beyond the obvious conflict of interest, however, this model has a more fundamental flaw: it is not designed to detect systemic risk. That weakness was laid bare during the 2008 global financial crisis, which is estimated to have cost the US economy $22 trillion and prompted the Dodd-Frank reforms, including the creation of the Financial Stability Oversight Council.

 

A FINRA-style model is ill-suited to AI oversight. Instead, governments should build their own AI expertise, drawing on the public oversight mechanisms introduced after the emergence of generative AI. One such measure was former US President Joe Biden’s executive order requiring developers of frontier AI models to share the results of their safety tests with the federal government before release. That same year, UK Prime Minister Rishi Sunak launched the AI Safety Summit, commissioned the inaugural International AI Safety Report, and paved the way for the establishment of the AI Safety Institutes.

 

Following the unveiling of Anthropic’s Mythos—the most powerful frontier model to date—Trump issued his own executive order. But it falls short, leaving pre-release evaluation voluntary. That creates a profound imbalance: today’s increasingly powerful AI systems require independent public institutions with the expertise and authority to oversee them, and Trump’s executive order provides neither.

 

The same asymmetry is evident in how governments approach AI risks. They tend to treat AI primarily as a national-security challenge, while devoting far fewer resources to addressing its effects on employment, health care, education, and other aspects of everyday life. Even the European Union’s AI Act, despite requiring assessments of such societal risks, places greater emphasis on cybersecurity.

 

One notable exception is the growing recognition of the risks AI poses to children online, which has led to legislative and judicial action in many countries, including the US. But its broader societal consequences—such as its effects on cognitive development, women’s economic opportunities, and privacy—have yet to prompt a coherent strategic response.

 

These skewed priorities can be traced to the framing of AI development as a geopolitical race. But technological leadership means little if it comes at the expense of workers’ health, social resilience, and people’s ability to lead meaningful lives. A country that develops the world’s most capable AI models while leaving its workforce deskilled and its public discourse polluted by misinformation has not won anything worth winning.

 

Addressing these societal challenges is not at odds with competitiveness. On the contrary, it is essential to both competitiveness and national security. Ultimately, what determines success in the age of AI is not just the capabilities of a country’s frontier models, but also the strength of the institutions that protect its citizens’ welfare: schools and universities, health and social care, and an independent judiciary.

 

Yet those are precisely the institutions struggling to keep pace with technological change. In education and labor markets, the OECD has found that AI’s benefits are accruing unevenly, reinforcing existing inequalities: Students who are already advantaged are better positioned to benefit from AI-powered learning, while skilled workers adapt more easily than those with fewer qualifications.

 

Similar trends can be seen in health care, social welfare, and the justice system. As decision-making is delegated to AI at scale, mistakes become harder to challenge, especially for those least able to do so. An AI-hallucination database maintained by Damien Charlotin of HEC Paris has documented more than 1,800 cases worldwide in which judges identified fabricated AI-generated material, often submitted by self-represented litigants who could not afford legal counsel.

 

The implications for judicial integrity are far-reaching. Courts depend on the ability to verify facts and evidence, but AI is making that task increasingly difficult, with the burden falling disproportionately on those already at a disadvantage. In July, for example, India’s Supreme Court set aside two tribunal rulings based on non-existent case law, warning that reliance on unverified AI-generated precedents “subverts the rule of law.”

 

To be sure, many of AI’s societal effects cannot be evaluated before deployment in the same way as cybersecurity risks. But that is all the more reason for continuous public oversight.

 

Instead of another industry-run agency, what the US and other countries need is the AI equivalent of the Financial Stability Oversight Council: a regulatory body charged with identifying systemic risks to education, health, labor markets, and the judiciary. We have already established such bodies to safeguard financial stability and national security. It is time to do the same for the institutions on which society depends.

 

Gabriela Ramos, co-chair of the Task Force on Inequalities and Social-Related Financial Disclosures, is a former assistant director-general for social and human sciences at UNESCO, where she oversaw the development of the Recommendation on the Ethics of AI, and a former OECD chief of staff and sherpa to the G-20, G-7, and APEC. Emilija Stojmenova Duh, associate professor of Electrical Engineering at the University of Ljubljana, is a member of the Globethics Board of Foundation and a former minister of digital transformation of Slovenia.