Last month, the artificial intelligence company Anthropic announced a model called Mythos that can find and exploit software vulnerabilities at machine speed. The system identified thousands of previously unknown security flaws in every major operating system and every major web browser, including a 27-year-old bug in OpenBSD that no human researcher had ever spotted. In the wrong hands, a tool like this could devastate American banks, hospitals, power grids, and military networks.
Anthropic chose not to release Mythos publicly. The company instead selected roughly 40 trusted partners and is letting them use the model to patch software before adversaries can exploit it.
Americans should be glad Anthropic made this call. They should also welcome the news that the Trump administration is now considering a formal review process for the most powerful AI models before public release. Reporting from The New York Times this week indicates the White House is drafting an executive order to create a working group on AI oversight, with senior officials including Chief of Staff Susie Wiles and Treasury Secretary Scott Bessent shaping the approach. This is the right move at the right time, and it builds directly on the foundation President Trump has already laid.
From his first day in office, the president recognized that artificial intelligence is a strategic asset. Executive Order 14179, signed in January 2025, removed Biden-era barriers to American AI development and established the policy of global AI dominance. The administration’s July 2025 AI Action Plan, titled “Winning the Race,” laid out a comprehensive strategy across innovation, infrastructure, and international diplomacy. In December 2025, the president signed another executive order preempting the patchwork of state AI laws that threatened to bury American firms in compliance burdens.
Each step strengthens America’s position, and the next logical step is to ensure that the most powerful systems emerging from American labs continue to serve American interests.
The numbers show why this works. The United States invested $285.9 billion in private AI capital in 2025, more than 23 times China’s $12.4 billion. American institutions produced 50 of the world’s top AI models last year, compared to 30 from China. The U.S. funded 1,953 new AI companies in 2025, more than 10 times any other country. America hosts more than 4,000 data centers, roughly 10 times China’s count. Every major frontier AI laboratory in the world is American.
The Mythos episode shows what happens without federal oversight. A private company in San Francisco made a national security judgment that will affect every American who uses a computer. Anthropic chose responsibly, but the next firm facing the same decision may not. Even when private executives make the right call, they cannot substitute for politically accountable institutions. Voters can hold a president accountable and pressure their senators, but they cannot vote out a San Francisco chief executive who decides which AI capabilities the country gets.
A formal review process puts these decisions where they belong, in the hands of officials who answer to the American people. The framework should remain lean, focused, and explicitly tied to national security. Federal agencies already equipped for this work, principally the National Institute of Standards and Technology and a small number of intelligence and cyber components, can evaluate frontier-class models against objective capability thresholds before public release. The vast majority of AI applications, from medical imaging to fraud detection to consumer chatbots, would face zero new requirements.
Critics will argue any oversight slows American innovation and helps Beijing. The data refutes them. China is not unregulated. The Cyberspace Administration of China requires generative AI services to obtain government approval before public release. Beijing already does this work. The choice is between American policymakers designing rules that serve American interests or leaving the field to commercial incentives that do not account for national security.
The president understands the stakes. He treats AI dominance as central to economic competitiveness and national security. A targeted, lean review process for frontier-class models extends that framework rather than retreating from it. Such a process denies adversaries the capability gains they would otherwise harvest from open-weight releases, protects the export control regime the administration built, and returns sovereign decisions about national security to sovereign hands.
America is winning the AI race because of bold private investment and confident federal leadership. The Trump administration’s willingness to evolve its approach as the technology evolves is the mark of strategic seriousness, not regulatory retreat. Done right, the next executive order will lock in American dominance for the long haul.
The president should act before the window closes.
