Can you indicate where in the article or the linked study those constraints are indicated? Military flights sims tend to be pretty decent for this.

The whole scenario was unrealistic. Simple rules of engagement, pre-programmed objectives, little or no EW, no ground control or AWACS, no ground defenses. The technology holds promise and will eventually be capable but we're still decades away from something that can replace human pilots across the full spectrum of missions.

The AI was defending, so it would've benefited from ground defenses, and the study indicates the human team had AWACS assistance.