Commands increasingly have access to software that flags service members as elevated risks for misconduct, insider threat, suicide, or security concerns, often by aggregating data and producing a score. When that score drives a real decision, such as a security-clearance suspension, removal from a sensitive duty, a command-directed evaluation, or referral for adverse action, the affected member naturally asks whether the use of an opaque algorithm can be challenged. The answer is a qualified yes. Due process applies, but how much process is owed, and what can be contested, depends entirely on what the score is used for.
Due process attaches to consequences, not to the tool itself
The Fifth Amendment’s Due Process Clause protects against government deprivation of life, liberty, or property without due process of law. The threshold question is never whether a commander used artificial intelligence; it is whether the resulting decision deprived the member of a protected interest. A risk score that merely informs a commander’s discretionary, internal judgment, with no adverse action attached, generally does not by itself trigger a due-process entitlement. The same score becomes contestable once it is the basis for tangible action.
That distinction sorts these disputes into two very different tracks: administrative actions, where civil due-process doctrine governs, and criminal prosecution at court-martial, where the more demanding constitutional and evidentiary protections of the military justice system apply.
In the administrative context: notice, response, and arbitrariness
Most behavioral-risk scoring drives administrative consequences, not criminal charges. There, the protections are those that attach to the specific interest at stake. A service member facing an adverse administrative separation, a clearance revocation, or a flag on assignments is typically entitled to notice of the basis for the action and a meaningful opportunity to respond. The contestable defect is usually not the existence of the algorithm but the secrecy around it: if the member is told only that a system rated him high-risk, with no access to the inputs, the logic, or the data, he cannot mount a meaningful rebuttal, and the process itself becomes suspect.
Civilian courts have wrestled with exactly this problem in the criminal-sentencing context. In State v. Loomis, 881 N.W.2d 749 (Wis. 2016), a defendant challenged a court’s reliance on a proprietary risk-assessment tool whose internal workings were a trade secret he could not examine. The Wisconsin Supreme Court upheld the use of the tool but imposed important guardrails: the score could not be the determinative factor in the decision, and decision-makers had to be given written cautions about the tool’s limitations, including questions about its accuracy and the populations it was validated on. The United States Supreme Court declined to take the case. Loomis is not military precedent and arose in civilian sentencing, but its reasoning is directly transferable to a command setting. An algorithmic score may inform a decision; it should not be the decision, and the inability to interrogate a black-box tool is a legitimate fairness objection.
The strongest administrative challenges therefore frame the issue as arbitrariness and lack of meaningful review: that the action rested on an unexplained, unvalidated, or untested score; that the member had no opportunity to confront the underlying data; that the inputs were inaccurate or stale; or that the tool was applied outside the population it was designed for. Where regulations require a decision to be supported by a rational basis or by specified findings, a bare algorithmic output that the command cannot explain may fail that standard.
In the criminal context: confrontation, disclosure, and expert assistance
If a risk score is offered to support or shape a court-martial, the protections are far more robust. The accused has a constitutional right to confront the evidence against him and a right to a fair trial. Several familiar levers apply.
Authentication and reliability. Output of a software system must be authenticated, and where its accuracy is contested the proponent must show the system produces a reliable result. A score whose methodology cannot be explained or validated is vulnerable to exclusion.
Confrontation and the human-versus-machine problem. The accused is entitled to confront the witnesses against him. A score generated by proprietary software, presented without a witness who can explain and be cross-examined on its workings, raises a serious confrontation problem. The defense can demand a sponsoring expert and can attack the inability to cross-examine an algorithm.
Discovery and expert assistance. The defense is entitled to discovery of material evidence and, on a proper showing, to expert assistance. Where the government relies on a risk tool, the defense can seek disclosure of the algorithm’s inputs, validation studies, error rates, and source documentation, and can request a defense expert to evaluate it. A refusal to disclose enough to permit a meaningful challenge can itself be a due-process violation.
Relevance and unfair prejudice. Even an admissible score must survive the balancing test that excludes evidence whose probative value is substantially outweighed by the danger of unfair prejudice. A veneer of mathematical certainty over a tool with a poor or unknown error rate is precisely the kind of misleading aura that balancing is meant to police.
What is actually contestable
Pulling these threads together, a member can realistically contest several things: the accuracy and completeness of the data fed into the tool; the validity and error rate of the algorithm; the secrecy that prevents meaningful rebuttal; whether the score was treated as determinative rather than as one input among many; whether the member received notice and an opportunity to respond proportionate to the interest at stake; and, in the criminal forum, whether the government has laid an adequate foundation and permitted confrontation. What is generally not contestable is the bare fact that a commander considered an algorithmic input in the exercise of legitimate discretion, so long as the member’s protected interests were respected and the score did not stand in for an actual, reviewable decision.
Practical posture for the affected member
The member’s first move is to demand the basis for the action in writing and to request the underlying data and the nature of the tool. That request both creates a record and tests whether the command can articulate a rational, individualized justification beyond the score. If the consequence is administrative, the response should attack inaccurate inputs and the absence of meaningful review. If charges are referred, counsel should treat the tool as contested forensic evidence and litigate authentication, confrontation, discovery, and unfair prejudice. The unifying principle is simple: due process is indifferent to whether a human or a machine flagged the member, but it is not indifferent to a deprivation imposed on the strength of a score the member was never allowed to see or challenge.
Disclaimer
This article is provided strictly for general educational and informational purposes. It is intended to explain how the Uniform Code of Military Justice (UCMJ), the Rules for Courts-Martial, the Military Rules of Evidence, and related military administrative processes work as a matter of public legal education. It does not constitute legal advice, a legal opinion, or a recommendation about any particular case, and it is not a substitute for advice from a qualified military defense attorney who can evaluate the specific facts and command, service, and jurisdictional circumstances involved.
Reading this article, or contacting any website on which it appears, does not create an attorney-client relationship between the reader and any law firm, attorney, or author. Every court-martial, nonjudicial punishment action, administrative separation, and security-clearance matter turns on its own facts, the charged articles, the convening authority, the service branch, and the evidence, and outcomes vary widely from one case to another.
Military law also changes over time. The Military Justice Act of 2016 (effective January 1, 2019) and subsequent National Defense Authorization Acts renumbered and rewrote many punitive articles, revised the Article 32 preliminary hearing, and altered sentencing, clemency, and appellate procedures. Statutes, regulations, executive orders, the Manual for Courts-Martial, and decisions of the service Courts of Criminal Appeals and the Court of Appeals for the Armed Forces may have been amended, superseded, or reinterpreted after this article was written, and article numbers or procedures cited here may have changed.
For these reasons, no reader should act or decline to act based on this content without first consulting a licensed attorney experienced in military justice about their own situation. The author and publisher make no warranty, express or implied, as to the accuracy, completeness, timeliness, or current applicability of the information provided, and disclaim any liability for any action taken or not taken in reliance on it. If you are facing investigation, charges, or an adverse administrative action, time limits may apply, and you should seek qualified counsel promptly.