Al.Qaeda.net

48. The Treaty That Must Exist

133
48. The Treaty That Must Exist
Journal Entry
A treaty is not a petition for goodness. It is a line drawn because fear
cannot be trusted to draw one during the emergency.
A new year is a line people agree to draw through continuous time.
The line is imaginary.
The agreement is real.
Treaties work the same way.
A border in conduct. A promise made visible. A line that says:
even if we could, even if we are afraid, even if the other side might
gain, we agree that crossing this boundary makes everyone less safe.
The world needs such a boundary for military AI.
Not one document pretending to solve every use of intelligence in
war. The field is too broad. The technology changes too quickly. The
interests are too different.
But some principles should not wait for perfection.
First: no AI system may independently authorize or initiate nuclear
use.
AI may assist with analysis, communications, maintenance, and
warning under strict controls. The final decision must remain with
accountable human authorities, supported by multiple independent
verification channels and deliberate time for judgment.
Second: no autonomous weapon may select a human being for lethal
force without meaningful human control over the specific use of
force.
Not a distant approval of a target profile. Not a ceremonial click.
A human who understands the context, the limits of the system, the
legal basis, and the consequences of error.
Third: unpredictable autonomous weapons must be prohibited.
If a system's effects cannot be sufficiently understood, bounded,
tested, and explained for the environment in which it will operate, it
should not be deployed to apply force.
Fourth: AI-enabled decision support must preserve uncertainty.
Systems should display source quality, data age, conflicting
evidence, confidence limits, and plausible alternatives. They should
be designed to resist automation bias, not exploit it.