AXY News Global

AXY News Global · defense

Army Cyber Command Tests When AI Agents Can Work Without Humans

Lt. Gen. Christopher Eubank says the command is learning, task by task, when AI agents can act alone and when people must stay in the loop

By AXY Defense Desk ·

Reporting via Military Times
Reporting via Military Times (Military Times)

Key points

The head of U.S. Army Cyber Command says the service is continuously observing and adjusting how people oversee artificial intelligence, and that it may eventually let computer agents carry out some work on their own. Lt. Gen. Christopher Eubank made the remarks at the Association of the United States Army conference in Washington, according to a Military Times report published on October 19, 2026. His account offers a detailed look at how a U.S. military cyber organization is deciding how much freedom to give AI systems, and where human judgment must remain part of the process.

The central message is that the Army has not yet handed autonomy to AI agents in its cyber work. Eubank described a process in which military personnel are learning in real time where to draw the line between letting software act and keeping people in control. That process, he suggested, will determine which tasks the command eventually allows machines to handle without human intervention.

What Eubank said about the process

Eubank described an iterative, human-in-the-loop process. In that model, a person stays involved in the system’s work and can review, approve, correct or stop what the software does. Military personnel, he said, learn in real time how to balance giving AI agents autonomy against exercising control over them. The process is continuous rather than a single decision made at one point in time.

In his words, the work is happening “daily and weekly inside of the headquarters as we watch agents learn and grow, and then determine what we should let them do on their own versus have a human in the loop,” Eubank said. He added: “All work right now is safeguarded human-in-the-loop as we continue to learn across this very important emergent technology.”

The two statements together set out the current position. The command is observing agents, assessing what they can do and deciding case by case what they may be allowed to do unsupervised. Until those decisions are made, the work is handled under human supervision. Eubank did not say the Army has approved any agent to operate independently, and the source material does not identify any task that has been cleared for unsupervised operation.

Eubank’s phrase “emergent technology” also signals that the Army regards agentic AI as still taking shape in military use. Leaders are not treating the current rules as settled policy for the long term. They are treating them as a working arrangement that will be adjusted as experience accumulates.

How the oversight works day to day

According to the report, military personnel currently watch AI agents perform routine tasks. A governance board within Army Cyber Command decides whether a given agent should be allowed to keep performing a task without human intervention, or whether it should be permitted to work unsupervised. In that arrangement, the board acts as a checkpoint between observation and autonomy.

During the current iteration, every machine AI team checked in with its human team daily and was not operating alone. The reporting describes this as a daily reporting rhythm, and it is one of the most concrete safeguards mentioned in Eubank’s account. It means that no agent is described as working in isolation during this phase.

Eubank also said that every AI agent was assigned the same work role as a human, and that the command trained those agents to become masters of their craft in the same way the Army trains its people. Giving an agent a defined role places it inside a structure of responsibility similar to the one that governs soldiers and civilians. It has a defined function, a supervising team and a standard against which its work can be judged. The source does not explain how those standards are written or measured.

The source material also does not describe the board’s membership, how often it meets, or what evidence it requires before approving autonomy for a task. Those details would help readers judge how strict the process is, but they should not be assumed.

Why speed is driving the debate

Eubank said the pace of change has been dramatic. In two days, he said, an AI agent could complete work that took an analytics support officer six months. The comparison is vivid, but it is a single example given in a conference setting, not a formal benchmark of the command’s output. Its value lies in showing why senior leaders see the question as urgent.

Speed cuts both ways. A tool that compresses months of analysis into days can free people to work on harder problems. The same compression, however, means that a flawed output can spread or be acted on before a human reviewer has time to examine it. That is the logic behind keeping review steps in place while the Army learns how agents behave over time.

The tension is familiar in other industries that have automated complex work. Organizations generally find that the faster a system runs, the more important it becomes to define checkpoints, logs and rollback procedures. Eubank’s remarks suggest the Army is applying that same reasoning to military cyber operations, although the source material does not describe the specific controls the command uses.

What Army Cyber Command uses AI for

According to Eubank, the command uses artificial intelligence to operate, secure, defend and attack networks, and to make sure the technology works hand in hand with the human workforce. He also said AI has enabled the command to move faster than it has ever moved and to sift through large volumes of data at machine speed. The stated benefit is that people can then focus on solving critical problems rather than on sorting information.

Cyber operations cover several distinct activities. Operating networks means keeping communications and systems running. Securing and defending networks involves detecting and stopping intrusions by hostile actors. Offensive cyber activity, often called attack in military language, involves actions taken against an adversary’s systems, and it is subject to legal and policy limits set by national authorities. Eubank’s list spans all of these areas, which is part of why the question of autonomy is sensitive. Software that acts quickly in one area may face very different rules in another.

When an agent carries out a sequence of steps without a person approving each one, the command must be able to reconstruct what it did and why. The source material does not describe how the command records agent actions or how its audit trails are kept. Still, the human-in-the-loop model can be understood as a way of keeping that accountability intact while the technology is still being evaluated.

Background: how the military has approached autonomy

The principle of keeping humans involved in consequential machine decisions predates the current wave of AI tools. The Pentagon’s directive on autonomy in weapon systems, first issued in 2012, requires that autonomous and semi-autonomous weapons be designed to allow commanders and operators to exercise appropriate levels of human judgment over the use of force. That policy was written mainly for weapons, but it has helped shape wider debate about how automated systems should be governed across the military.

Agentic AI is a newer term. A conventional AI tool typically answers a question or produces an output that a person then uses. An agent, by contrast, is designed to pursue a goal across several steps, which can include using other software, making changes in a system or passing work from one task to another. In plain terms, an agent may do more than recommend an action; it may carry out work that a person would otherwise have performed step by step. That shift is why oversight becomes a practical matter of permissions, monitoring and approval points, rather than a simple question of whether someone reads the final result.

Human-in-the-loop is one of several oversight models used in discussions of automated systems. In broad terms, human-in-the-loop means a person must take part in or approve each significant action. Human-on-the-loop generally means people monitor the system and can intervene, while human-out-of-the-loop means the system acts without real-time human involvement. Eubank’s description places the Army’s current practice at the most closely supervised end of that range, with autonomy for specific tasks being considered as the command gains experience.

Why it matters to service members and the public

For service members, the process could change how work is divided between people and software. If agents are allowed to handle routine analysis or network tasks without supervision, human staff may spend more time on judgment-heavy decisions. The source material does not describe any planned changes to personnel or job structures, so any such outcome remains speculative. Still, people whose work overlaps with agent tasks have a direct interest in which tasks are cleared and how their supervisory roles may change.

For the public, the stakes concern reliability and control. Military networks carry sensitive information, and cyber operations can affect systems far beyond their intended target, sometimes in ways that are hard to reverse. Analysts expect that the rules the Army settles on for autonomy in cyber work could influence how other military branches and government agencies approach similar questions, although the source material does not cite any formal guidance that would bind them.

The debate also reflects a broader policy question about the pace of adoption. Faster tools promise gains in capacity, but governments and militaries must decide how much verification to demand before trusting software with tasks that once required trained people. Eubank’s remarks suggest the Army is choosing a stepwise path, testing autonomy in small increments rather than making a rapid handover of responsibility.

What to watch next

The most important open question is which tasks the governance board will eventually allow agents to perform without human intervention. Eubank did not name specific tasks or a timeline for that decision, so any expansion of autonomy would be a new development to track.

Observers will also want to know whether the Army publishes criteria for how the board judges an agent’s performance, and how errors will be handled when a system works without supervision. Clear answers would show whether the human-in-the-loop approach is treated as a temporary stage or as a lasting principle of military cyber work.

It will also be worth watching whether the approach spreads beyond Army Cyber Command, and whether Congress or Pentagon leaders set broader standards for AI agents in military operations. For now, the clearest message from Eubank’s remarks is that the command is proceeding carefully, testing autonomy one step at a time while keeping people responsible for the work.

Questions and answers

What did Lt. Gen. Christopher Eubank say about AI in the Army?

Eubank said Army Cyber Command is constantly observing and refining human oversight of artificial intelligence, with the possibility of releasing some AI agents to work on their own. He described all current work as safeguarded by human oversight.

What does human-in-the-loop mean for military AI?

Human-in-the-loop means a person stays involved in an automated system’s decisions or actions. In this case, a governance board within Army Cyber Command decides whether an AI agent can keep working without human intervention.

Will Army AI agents work without humans?

Not for current work, according to Eubank, who said all work right now is safeguarded human-in-the-loop. He did not name specific tasks or a timeline for any unsupervised operation.

What does Army Cyber Command use AI for?

According to Eubank, the command uses AI to operate, secure, defend and attack networks, and to sift through large amounts of data at machine speed so people can focus on critical problems.

Topics: Army Cyber Command, Christopher Eubank, artificial intelligence, human-in-the-loop, AI agents, military cyber operations, US Army

Sources: Military Times

https://axynewsglobal.com/article/army-cyber-command-tests-when-ai-agents-can-work-without-humans