About $1.48M to Penta Research — the Missile Defense Agency studying AI in scenario planning and design (a number beginning HQ086)

1The subject is the assumptions themselves

What the description names is scenario planning and design: the work of building the storylines that set out what might happen.

Training and evaluation alike begin by placing assumptions and measuring judgment or performance on top of them. Skew the assumptions and the conclusions drawn skew with them. This research reads as applying AI not to the side that produces results but to the side that builds the ground on which results are measured.

2What it means for a machine to write the assumptions

When people build assumptions, they are pulled toward the range of their experience. Have a machine build them and combinations nobody thought of may appear, while what lies outside the learned range stays absent just the same.

Both carry limits, so the question of how to verify the coverage of a set of assumptions remains. The record does not describe methods, so how this project answers that cannot be read. What it supports is the frame: research and development addressing scenario planning and design.

3A contract number beginning HQ

Across the 222 federal spending records this site holds as of 2026-09-01, all 155 beginning with a letter were awarded by the Department of Defense. Seventeen begin with HQ, covering Washington Headquarters Services and the Missile Defense Agency.

This contract number begins HQ086, corresponding to the Missile Defense Agency. Within the same HQ the following digits separate the buyer, so the first two characters alone do not pin down the subordinate body. The scheme is layered, and how far it is read changes how fine the information gets. The period runs exactly two years, from 2023-03-14 to 2025-03-14.

4Who writes the scenarios

Having a machine write the scenarios carries limits different from those of a person writing them.

In training and in evaluation alike, a scenario is set first and judgement or performance measured on it. Where the scenario is skewed, so is the conclusion drawn from it. This research reads as putting AI on the side that builds the ground for measurement rather than the side producing results. The record states no method, so how the plan answers that question cannot be read.