On the first day of the war on Iran, February 28, Palantir’s Maven Smart System steered the U.S. military to strike at more than 1,000 targets. One was a primary school in Minab, where more than 160 people were killed, most of them children. Al Jazeera examined that strike this week under the question of whether AI systems can make moral decisions in war.

Oof. No. The question the article debates is undecidable, and therefore a useless distraction. The question it should be asking instead has an answer from 1931.

Note, for example, how the article quotes the CENTCOM commander on the safeguard:

Humans will always make final decisions on what to shoot and what not to shoot and when to shoot, but advanced AI tools can turn processes that used to take hours and sometimes even days into seconds.

The claim that the decision stays human fails scrutiny. Operator fatigue is the obvious counter point. A flood of rapid decisions leads to inability to assess the choices. When the time to decide shrinks to seconds, repeatedly, the decisions stop being human. The claim from CENTCOM, about itself, fails to hold ground when evaluated from outside.

Kurt Gödel in 1931 famously confronted exactly this form of claim. His paper constructed, inside a formal system, a sentence about that system’s own capacity to prove things:

We are therefore confronted with a proposition which asserts its own unprovability.

(Kurt Gödel, Über formal unentscheidbare Sätze der Principia Mathematica und verwandter Systeme, 1931)

Proposition XI followed. A consistent system rich enough for arithmetic places its own consistency beyond its own proof. Verification requires a reader standing outside the system. Every statement the system makes about its own soundness, however elaborate, carries the same evidential weight from inside: zero.

The two loops that military and industrial doctrine run on were built before this became an engineering problem. In an OODA loop, the pilot reports on an adversary the report describes from outside. In the PDCA, a Check worked because others stood in the room with the operators to challenge the operator’s account of their own work.

Maven refuses outside evaluation. The pipeline runs find, fix, track, target, engage, assess as a single chain. The system that generates the target assesses the strike. The legal sign-offs travel inside the pipeline. Civilian harm review now runs on a tool built on Maven itself, cut to nine staff from 40. Every phase of the loop reports on itself, and Gödel proved the limited utility of this ninety-five years ago.

The MOCA loop, Gebrüder Ottenheimer Brief №7, is built to investigate what Gödel identified. Verification of AI lives outside the chain that proposed the action. Here are the four phases against Maven:

MOCA	Maven at Minab
Modify. The operator’s stated change lands in an immutable log before any agent runs. The original intent is auditable.	The targeting basis lives inside the pipeline. Four months on, the error surfaced through Bloomberg reporting on outdated satellite intelligence, and the record remains contested.
Observe. Many narrow agents, each isolated, each partial. Authority over the full picture belongs to the diversity in the log.	One fusion system merges more than 150 sources into a single authoritative picture for 20,000 operators.
Converge. Proposals and reports meet in a log the agents cannot edit. Contradictions surface there.	Battle damage assessment runs inside the pipeline that generated the target. Civilian harm review runs on a Maven-built tool.
Act. The commit gate belongs to an operator outside the chain that proposed the action.	The human gate remains in name only while the timeline compresses from hours to seconds.

Al Jazeera should not ask whether Maven can carry moral weight, given the logic. Gödel points at a smaller and harder question: who outside CENTCOM can replay the Minab targeting chain and detect tampering?

Four months of a stalled Pentagon investigation were avoidable. An outside log, with outside analysis, would have exposed the record in February.