Al.Qaeda.net

Sensors fail. Adversaries manipulate signatures. Analysts work under

57
Sensors fail. Adversaries manipulate signatures. Analysts work under
pressure. The consequences of a false positive are not a mistaken
advertisement.
The ICRC's concern is not theoretical: when systems select and
engage targets based on sensor data and generalized target profiles,
human users may not know the specific person, time, or place in
advance. [7]
The civilian in the dataset needs legal protection before the data
becomes a target package.
Not only privacy law.
War-specific protection.
Data provenance.
Confidence calibration.
Freshness requirements.
Independent corroboration.
Mandatory review of civilian-harm risk.
Audit logs that preserve who knew what and when.
A way to learn from error without hiding it behind classification.
The phrase "collateral damage" is a linguistic machine. It converts
a person into a secondary effect.
AI can intensify that distance by making the decision look
analytical.
A colored box around a building.
A score beside a name.
A route predicted across a map.
The interface can be technically accurate and morally numbing.
Design matters.
A system should display uncertainty, conflicting evidence, data
age, civilian presence, and the consequences of being wrong. It should
not make irreversible action the easiest button.
The civilian in the dataset is not an edge case.
The civilian is the reason the law of war exists.
The machine must not be trained only to find what can be struck.
It must be designed to show why a strike should not occur.
A PROFILE IS NOT A PERSON.
FFTAC ACTION - Find one category that reduces a person to risk,
eligibility, or target status. Demand the evidence, the error rate, human
review, and a path to correction.