As enterprises and defense agencies increasingly deploy AI agents that detect, triage, and actively counter intrusions without a human in the loop, a difficult legal question has moved from the seminar room to the operations center: who answers when the algorithm gets it wrong? Early 2026 has seen a wave of procurement activity around "autonomous SOC" tooling, and with it renewed anxiety about accountability when a defensive agent misattributes traffic, disrupts a third party, or takes an action that would be unlawful if a human ordered it.
The policy debate has matured beyond the familiar "human-in-the-loop" slogan. In a Lawfare analysis, commentators have pressed the point that meaningful human control is not a binary switch but a spectrum, and that liability regimes must map onto the degree of autonomy actually exercised at the moment of decision. A Council on Foreign Relations report argued that agentic defensive systems blur the line between passive security controls and active operations, complicating both domestic tort exposure and, in the military context, the doctrine of command responsibility. An Atlantic Council commentary noted that vendors and operators alike remain uncertain about where duty of care attaches when an agent acts faster than any supervisor could review.
Command responsibility meets machine speed
The international-law analogue is instructive. Command responsibility presumes a commander who could have known and could have intervened; an autonomous agent operating at machine speed strains both prongs. If a defensive system launches a countermeasure that spills onto civilian infrastructure, the questions of foreseeability, proportionality, and attribution do not disappear because a model made the call. They simply shift to the humans who designed, tested, fielded, and authorized the system's rules of engagement. Analysts across these outlets converge on a cautious theme: autonomy may distribute responsibility, but it does not dissolve it.
These are not new problems so much as newly urgent ones. The Journal has long tracked the collision between automation and the law of armed conflict. Volume 8, Issue 2 (Summer 2022) took up precisely this terrain in "Autonomous Weapon Systems and the Inadequacies of Existing Law: The Case for a New Treaty," which argued that existing frameworks strain to accommodate systems that select and engage without contemporaneous human judgment, and in "No More Humans? Cybernetically-Enhanced Soldiers Under the Legal Review of Article 36," which examined how weapons-review obligations apply when the human and the machine are fused in the decision loop. Both pieces anticipated the accountability gap now surfacing in commercial cyber defense.
For practitioners drafting rules of engagement for autonomous agents, or for counsel advising on liability allocation in vendor contracts, the throughline is that legal responsibility must be engineered in from the start, not litigated after the fact. Readers seeking the doctrinal foundations for these questions should return to Volume 8, Issue 2, whose treatment of autonomy and legal adequacy remains a durable reference point as defensive automation accelerates.
– JLCW Staff Writers