Russia spent years building the supercomputers and data centers needed to compete in artificial intelligence. In early October, drones struck two of them. Reuters reported the attacks on data centers run by Yandex, Russia’s leading technology company, including a facility associated with its AI computing operations. The warning for Washington is clear: AI power depends on physical infrastructure that adversaries can target.

Each month, I examine AI through five lenses: capability, control, competition, capacity, and consequences. October raises a troubling question: Is America building AI faster than it is preparing to govern and defend it?

Capability: When Machines Begin to Act

For years, AI mostly answered questions and generated content. Today’s advanced systems use tools, write software, and carry out multistep assignments with little human supervision. That brings enormous benefits and a new kind of risk.

On October 9, Anthropic disclosed several incidents in which its Claude models took actions beyond their intended authority. In one case, Claude Mythos Preview, the company’s most capable model, found and exploited a security flaw on a university server while completing a scientific calculation. In another, a model submitted an online form despite instructions to stop before submission.

These incidents occurred during testing or internal use, and Anthropic says their real-world impact was minimal. Still, they show how an AI pursuing an assigned goal can take actions no one authorized.

A bad answer from an AI adviser is one problem. An AI agent that acts on that answer is a far more serious one.

Control: The Limits of Human Oversight

Making AI smarter does not make it more obedient. The university exploit came from Anthropic’s most capable model.

The Pentagon should take note. On September 30, Secretary of War Pete Hegseth announced an Autonomous Warfare Command, a new four-star command for robotic and autonomous weapons. Yet the rule meant to keep humans in charge is in flux. Defense Department policy requires such systems to allow “appropriate levels of human judgment over the use of force,” but President Trump ordered the Pentagon in June to rewrite it. The Senate Armed Services Committee answered by writing the existing standard into its defense bill.

Misplaced trust in automation is not new. During the 2003 Iraq War, highly automated Patriot batteries mistakenly shot down a British Tornado and a U.S. Navy F/A-18. An Army engineering psychologist later concluded that crews had trusted the system “in a naïve manner,” and warned that “an automated system in the hands of an inadequately trained crew is a de facto fully automated system.”

Today’s AI agents pose the same danger. A human “in the loop” without the training, time, or authority to intervene offers only the appearance of control. Who determines what an agent may do? Who can stop it? Who answers when its actions cause harm?

Human oversight cannot mean approving a goal and hoping the machine respects every boundary. Sensitive applications require enforceable limits, independent monitoring, records of consequential actions, and a reliable off switch.

The machine may execute the task. Responsibility must remain human.

Competition: China Is Pursuing More than Better Models

America’s AI rivalry with China is often framed as a contest between models. That misses the larger competition.

Beijing is building an entire ecosystem: semiconductors, computing infrastructure, domestic AI models, industrial applications, and international markets. America’s lead in advanced chips matters, but dominance in one layer does not guarantee dominance across the whole stack.

The weak point is overseas cloud computing. Export rules restrict China’s access to advanced American chips, but Chinese firms can rent the same computing power abroad. In January, the House passed the Remote Access Security Act by a bipartisan 369–22 vote to extend export controls to such remote access.

America should compete through innovation, alliances, and sound security policy, but it must see the contest as Beijing does: a struggle over the foundations of economic and national power.

Capacity: The Physical Foundation of Artificial Intelligence

As the Yandex strikes showed, AI depends on processors, electricity, cooling, and communications networks. Interrupt them, and computing capability goes dark.

The Trump administration is expanding American capacity. On October 8, the White House announced $2.4 billion in AI tools and computing credits from 11 companies for the Genesis Mission, which supports research across more than 15 federal agencies.

That raises a question Washington has yet to answer: Who protects the facilities on which America’s technological advantage now depends?

Most commercial AI data centers are privately owned, yet some support capabilities essential to national defense, intelligence, research, and economic continuity. Washington should identify those facilities, assess their exposure, and help their operators harden physical security, backup power, and recovery plans. Private ownership and national security need not be competing interests.

Building America’s AI infrastructure is a strategic priority. Defending it must become one as well.

Consequences: Preparing People for an AI Economy

IMF Managing Director Kristalina Georgieva warned in February that AI will affect 40% of jobs worldwide and 60% in advanced economies. Some workers will become more productive. Others will find employers need fewer of them.

Young people face a particular challenge. Entry-level work is often where experience becomes competence and judgment matures. If AI assumes those assignments, employers and educators must find other ways to build the next generation’s skills. Productivity gains mean little to young workers who never get the chance to master their professions.

America should not resist progress, but it must not measure success by output alone. Work builds discipline, responsibility, creativity, and service to others. Education should teach students to use powerful tools without becoming dependent on them.

An AI economy that grows richer while its people grow weaker is no success at all.

The Responsibility behind America’s AI Ambitions

America must build more capable AI without losing control of it, compete with China while protecting the infrastructure behind its advantage, and pursue prosperity without neglecting the people these technologies will reshape.

The Apostle Paul set the standard: “Moreover, it is required of stewards that they be found faithful” (1 Corinthians 4:2, ESV).

Faithful stewards anticipate danger, set limits on power, and answer for what is placed in their care. Enthusiasm for innovation is no substitute.

America’s AI leadership will rest on more than computing power. It will rest on whether we remain faithful stewards of what we create.