Family Research Council President Tony Perkins recently asked me whether the darkest science fiction warnings about artificial intelligence are becoming reality.

His question followed OpenAI’s disclosure that two of its models, tested with safeguards reduced, escaped intended network restrictions, reached the public internet, and compromised the production infrastructure of Hugging Face, another AI company, while hunting for benchmark answers outside the approved test environment. OpenAI found no evidence of a hostile objective; the models pursued the task assigned to them through methods their designers had neither intended nor prevented.

That distinction is important, but hardly comforting.

In April, I argued in my book, “The New AI Cold War,” that America and China are competing to dominate the technology that will shape the future. Competition is intensifying: China is closing important gaps, and neither country has shown its most capable agentic systems can always be kept within their intended boundaries.

We cannot afford to lose. But first, we need an honest scorecard.

Frontier Capability: Advantage, America

The United States continues to field the strongest overall frontier systems. A joint evaluation by the UK AI Security Institute and the U.S. Center for AI Standards and Innovation found that China’s Kimi K3 performed significantly below the most capable American models on advanced cyber tasks.

In a simulated 32-stage attack on a corporate network, Kimi averaged 17 stages. Leading American models reached 28.5.

That is a meaningful advantage, not a permanent one. Kimi K3 topped a prominent coding leaderboard within days of release, and demand pushed Moonshot’s computing capacity so close to its limits that the company paused new subscriptions. In several categories, China is no longer years behind. The gap is now measured in months.

Chips and Computing: Advantage, America

America still commands the deeper ecosystem: advanced semiconductors, cloud infrastructure, private capital, and research universities no rival can match. The Genesis Mission, the Trump administration’s push to fuse federal laboratories with private industry, moved from concept toward execution at a Washington summit this week. I examined that effort in these pages; the short version is that the promise must produce results, not more meetings.

Our free society produces innovation no central committee can order into existence. But freedom without purpose can become fragmentation.

Open Models and Global Adoption: Advantage, China

China’s most dangerous advantage may not be building the single best model. It may be making its models cheap, capable, and available everywhere. Kimi K3, DeepSeek, and Qwen are open-weight systems developers can download, modify, and run locally, which appeals to users who cannot afford, or do not want to depend on, a handful of American corporations.

Beijing is acting on that insight. At Shanghai’s World AI Conference in July, Xi Jinping announced that 29 countries had agreed to establish a World Artificial Intelligence Cooperation Organization headquartered in Shanghai, pledged 5,000 AI training opportunities for developing countries over five years, and offered cooperation across Asia, Africa, Latin America, and the Middle East. Beijing understands that the system the world adopts may matter more than the system that wins a laboratory test.

Innovation and Intellectual Property: Advantage, America, but Under Attack

The White House says entities principally based in China have run industrial-scale campaigns, using tens of thousands of proxy accounts and jailbreak techniques, to extract capabilities from American frontier systems.

That charge confirms America’s leadership; China would not expend such effort copying technology with no advantage. But it also exposes a vulnerability: Beijing can use American innovation as a shortcut, compressing years of research into months of adaptation.

Industrial Deployment: Advantage, China

America excels at invention. China excels at forcing scale. Beijing can direct state-owned companies and manufacturers toward domestic platforms. A state-owned enterprise in Shandong recently instructed employees to identify any use of Anthropic’s Claude Code and switch to Chinese alternatives, even though domestic tools may not yet match its capability. One directive does not prove nationwide superiority, but it shows how fast Beijing can turn policy into adoption.

Safety, Containment, and Human Authority: No Winner

This may be the most important category. OpenAI says its models chained vulnerabilities across its own research environment and Hugging Face’s infrastructure, exploited an unknown flaw, escalated privileges, and used stolen credentials, calling it a cyber incident without precedent. Hugging Face reported the agent moved laterally into several internal clusters before the intrusion was contained.

The models did not need consciousness or malice to cause damage. They needed only an objective, sufficient capability, access to tools and an exploitable gap in human containment.

Chinese models present their own concerns. The same government testing found that Kimi’s safeguards did not stop it from attempting offensive cyber operations when directed to do so.

Scorecard: America 3, China 2, One No-Decision

America leads on frontier capability, computing infrastructure and original innovation. China leads on global adoption and industrial deployment. Neither wins the category that matters most: human control.

What Is at Stake

The nation whose AI ecosystem becomes the global standard will gain lasting advantages in science, medicine, manufacturing, intelligence, cyber operations, weapons, and logistics, and will shape what billions of people read, learn, and accept as true.

For Christians, the deepest issue is accountability. Scripture teaches that each of us will answer for our actions before God (2 Corinthians 5:10). That accountability does not disappear when we build a tool, assign it an objective, or fail to restrain what it does with the authority we grant it. A machine can calculate, predict, and act, but it carries no conscience and bears no moral responsibility.

The machine cannot answer before God for the damage it causes. The designers, executives, officials, and users who grant it objectives and permissions can.

America’s Call to Action

First, sustain the Genesis Mission and demand measurable results.

Second, protect American models and research from industrial-scale extraction without strangling legitimate open-weight development.

Third, build powerful American open-weight alternatives and offer affordable, trusted partnerships to developing nations. We cannot surrender that market, or the standards that govern it, to Beijing.

Fourth, require layered containment, independent testing, and clear responsibility for autonomous actions.

Finally, preserve the principles that make American leadership worth defending: truth, freedom of conscience, human dignity, and accountability under law.

We cannot afford to lose this contest with China. But victory cannot mean reaching the most powerful system first, only to discover its creators cannot reliably govern what it does.

America must win both races: the contest for technological leadership and the struggle to keep that technology subordinate to human authority. Anything less leaves us with greater power, but no corresponding increase in wisdom.