Whelp, I guess artillery and grenades are fully autonomous.
It is much more useful to think about the reverse situation: how does viewing future autonomous weapons as a sort of artillery shell with high CEP (or whatever feature you imagine the future weapons has) clarify the morality of its use?
Am I correct in interpreting your sarcastic comment as meaning you think that pursuing some definition of "autonomous" is a nonstarter here?
The US military, like every other organisation outside the US, doesn't include conventional explosives under WMD.
If there's any point that a human cannot 'abort', then it needs to be classified as a WMD.
Out of human control I view swarm UCAVs as an unrecallable remote shotgun for the same reason as the article author, it's "inherently indiscriminate".
"The most widely used definition of "weapons of mass destruction" is that of nuclear, biological, or chemical weapons (NBC) although there is no treaty or customary international law that contains an authoritative definition. Instead, international law has been used with respect to the specific categories of weapons within WMD, and not to WMD as a whole."
However, there is one definition given in the Wikipedia article that seems to be sufficiently precise:
"""
(1) Any explosive, incendiary, poison gas, bomb, grenade, or rocket having a propellant charge of more than four ounces [113 g], missile having an explosive or incendiary charge of more than one-quarter ounce [7 g], or mine or device similar to the above.
(2) Poison gas.
(3) Any weapon involving a disease organism.
(4) Any weapon that is designed to release radiation at a level dangerous to human life.
"""
I definitely agree with your overall point that defining AFADS as a WMD will require careful navigation of the blurriness surrounding the category. On the other hand, it seems like that second definition is already in the territory of choosing arbitrary cutoffs (ie 113 grams).
An alternative perhaps is to focus on handling the "autonomous" part, rather than the "swarm" part, as the USMA article suggests. The definition here can be more precise/less arbitrary I think: a weapon that can at any time be overridden by an operator is not fully autonomous. However, if its regular operation includes a period where there is no operator in the position to override it (eg the drone(s) keep doing their thing while the operator is away from the controls), it would then count as fully autonomous. Having some authentication that the operator is at the controls at all times, available to override and legally responsible for the behavior of the drone(s) is the key to this approach I think.
[0] https://en.wikipedia.org/wiki/Weapon_of_mass_destruction