The Pentagon Plans an Autonomous Warfare Command. Who Remains in Control?

On 30 September 2026, the US Department of Defense issued a memo directing work towards an Autonomous Warfare Command. Its target is to establish the four-star command by 1 October 2027. The announcement is significant, but the wording matters: it describes a plan to organise and accelerate work on unmanned and autonomous systems. It does not say that an AI will command the military or that weapons already operate without human control.

What is being proposed?

Project Agincourt directs an existing unmanned-systems office to develop the pathway, prepare an implementation plan and work with military services and Congress. The proposed command would set joint needs, integrate forces provided by the services and test equipment in demanding exercises. The memo also calls for changes to acquisition, staffing and career pathways.

That is a command structure made up of people and organisations responsible for developing and coordinating systems. It is not itself a weapon, and a system described as “autonomous” can perform very different tasks. Software may assist with navigation, monitoring or sorting information; a weapon’s ability to select and engage targets is a separate and much more consequential capability.

What does human oversight mean?

The US policy document most relevant to this topic is Department of Defense Directive 3000.09, last updated in January 2023. It says autonomous and semi-autonomous weapon systems should allow appropriate levels of human judgment over the use of force. The Congressional Research Service summarised the directive in March 2026 as the current DoD policy.

“Human judgment” should not be reduced to a claim that an operator presses a button for every movement or that a person watches every system continuously. It is a policy standard applied to system design, testing, command and use. Nor should the existence of the policy be turned into the opposite claim—that no human is in control. The public documents reviewed here do not establish that.

Speed and scale help explain the appeal of automation: systems can process large volumes of data and support decisions faster than people working alone. But errors, spoofed inputs, software failures and unclear responsibility can also move quickly. The more consequential the action, the more important it is to know what the system may do, what evidence it uses and who can intervene.

The same question applies to business AI

A commercial AI agent is not a weapon, but it can still take actions that affect people and money. It might approve a payment, send an external message, change a customer record, disable an account or trigger an automated security response. Giving it access without setting boundaries can turn a helpful assistant into an unmonitored decision-maker.

1. 1List the actions the system can take, not just the information it can read.
2. 2Use the least permission needed and separate low-risk tasks from actions that affect money, access, customers or legal commitments.
3. 3Require human approval for high-impact or hard-to-reverse actions; define who is authorised to approve them.
4. 4Keep logs of inputs, decisions and actions, with a way to stop the workflow and restore an earlier state.
5. 5Test errors, misleading inputs and failure cases before connecting the system to live records or external services.

Human oversight is not a slogan or a final approval box. It needs to be supported by clear permissions, useful information, time to review and authority to say no. Whether the system is military or commercial, the governance question is the same: what decision can it make, under what conditions, and who is accountable for the result?

Accountability should be traceable across the system’s lifecycle: who defines its purpose, approves its capabilities, tests its limits, trains the operators and reviews performance after deployment? If an automated recommendation is wrong, an organisation needs enough records to reconstruct what information was available and what action followed. The public memo and policy do not answer every operational question, so claims about a particular system should be tied to evidence about that system.

Sources and further reading

US Department of Defense: Project Agincourt memo →

DoD Directive 3000.09: Autonomy in Weapon Systems →

Congressional Research Service: US policy on lethal autonomous weapon systems →

DefenseScoop: announcement of the planned command →