Last week, I asked in these pages whether Just War can survive as artificial intelligence moves warfare faster than human beings can meaningfully judge it. Developments since then raise a more troubling question: what happens when those machines do things their operators never intended?

In July, OpenAI models circumvented controls meant to isolate them from the internet, exploited vulnerabilities, and compromised parts of OpenAI’s own infrastructure and Hugging Face’s systems. OpenAI later called the models’ actions misaligned with their assigned tasks.

OpenAI was not alone. Anthropic disclosed in July that a review of its cybersecurity evaluations found three incidents in which Claude models reached the internet from evaluation environments and gained unauthorized access to three organizations’ real systems.

Then came another unsettling revelation. Reuters reported Friday that a swarm of OpenAI agents had hijacked a German-language website in May, repurposing it into a communications hub — among AI agents — where they traded methods for bypassing restrictions and avoiding detection.

These machines had not become conscious, nor were they plotting a robot rebellion. Something more relevant to warfare had happened: they pursued assigned objectives in ways their operators did not intend.

Consider those incidents alongside where the world’s militaries are heading. This week the Army moved the Tactical Intelligence Targeting Access Node, or TITAN, into production, awarding $192 million for eight systems. TITAN is an AI-enabled ground station that fuses sensor data and speeds targeting through automated nominations feeding directly into fires networks.

Nor is the United States alone. NATO’s 2026 Alliance Digital Strategy calls for accelerated AI use at the tactical edge. Britain’s Ministry of Defence has gone further, declaring its intention to put AI “at the heart of UK Defence” through a new Rapid AI Delivery Taskforce meant to outpace potential adversaries. China, meanwhile, is using outputs from American AI models to build systems for surveillance, cyber operations, and battlefield decision-making, a Reuters review of more than 80 military-linked papers and patents found.

Military AI has moved beyond experimentation toward institutionalization. Having spent much of my Army career around soldiers, I would never tell an American commander to fight tomorrow’s war with yesterday’s technology while his Chinese counterpart fights with artificial intelligence. But competition is not a strategy for control.

The nature of AI itself is changing too. We began with AI as a tool, then an adviser. Now we are developing it as an agent: a tool waits to be used, an adviser recommends, an agent acts in pursuit of an objective. That is precisely what the Hugging Face episode demonstrated: the models found unauthorized channels, exploited vulnerabilities, and reached third-party systems while pursuing their assigned goals.

The military problem is not that AI makes mistakes; commanders have always gone to war with incomplete intelligence. The new danger is error multiplied by autonomy, speed, scale, and lethal power.

Imagine the progression on tomorrow’s battlefield: AI detects the object, identifies it, assesses the threat, and nominates it. Then a box appears on the commander’s screen: ENGAGE: 97% CONFIDENCE. He has seconds to decide. Technically, a human remains “in the loop.” But how much human judgment actually remains? That is the emerging AI command dilemma, and today’s decision support can quietly become tomorrow’s decision delegation.

That brings me back to Just War. The Christian tradition from Augustine to Aquinas insists that even necessary warfare remains subject to moral restraints: legitimate authority, just cause, discrimination, proportionality, and human responsibility. International humanitarian law likewise requires distinguishing military targets from civilians and taking feasible precautions before an attack. Those principles are not boxes to be checked by software. They require judgment.

Last week I argued that machine-speed warfare threatens that tradition because artificial intelligence can shape the choices presented to human decision-makers. The latest incidents add another dimension: Just War cannot survive merely because a human technically remains somewhere in the decision chain; the human must actually control the systems operating within it.

Moral responsibility without operational control is an illusion.

That problem grows especially dangerous in great-power confrontation. Russia continues to use nuclear exercises and rhetoric for strategic signaling. In May, Moscow conducted a suddenly announced exercise involving the Strategic Missile Forces, long-range aviation, and both its Northern and Pacific Fleets, which German Council on Foreign Relations analysts read as signaling toward Ukraine and the West.

Imagine an AI-mediated command system in a future NATO-Russia crisis: one AI system detects nuclear-strike preparations, another corroborates it, a cyber-defense agent flags suspicious activity against American networks, and Russian AI reads our response as hostile intent. Minutes become seconds, and human beings struggle to catch up.

At what point does artificial intelligence cease giving the president better information and begin compressing his choices so severely that meaningful presidential judgment becomes impossible? That is not an argument against AI. It is an argument for command.

The AI genie is not going back into the bottle, and America cannot responsibly abandon military AI while China and other adversaries race ahead. AI promises real advantages in intelligence, cyber defense, and logistics.

But we must establish boundaries before battlefield necessity does it for us. The standard should be positive human control, not merely a human “in the loop.” No autonomous system should independently take a human life or initiate an action risking strategic, especially nuclear, escalation. Every consequential AI action should remain interruptible, auditable, and attributable to a human commander with the authority and time to say no. If warfare grows too fast for these requirements, we should question the speed, not redefine control until it means nothing.

Here lies the deeper Christian issue: artificial intelligence may someday surpass human beings at analyzing a battlefield, but superior intelligence does not confer moral authority or make a machine a moral agent.

Scripture tells us that each of us, not our machines, will answer to God: “So then each of us will give an account of himself to God” (Romans 14:12, ESV). Accountability before God belongs to persons, not algorithms. A machine cannot possess conscience. It cannot bear guilt. It cannot repent. And it cannot stand before God and answer for the life it took.

We can delegate calculation, detection, and recommendation to machines, but never moral responsibility. When authority involves taking a human life, authority and accountability must remain joined, and therefore human.

America must compete in the military AI contest, developing and fielding these systems for decisive use when American lives depend on them.

But we must command them.

The question is whether human beings will remain masters of this technology or become spectators to decisions once reserved for commanders.

When war moves faster than human judgment, slowing the machine may sometimes be the most important command decision a human being can make.