Wargaming suggests that instead of flying alongside manned aircraft, unmanned systems are most effective when decoupled. Their best roles are as long-endurance decoys to confuse incoming missiles or as autonomous swarms in a designated kill zone.

The most significant danger of autonomous weapons is not a single rogue robot, but the emergent, unpredictable behavior of competing AI systems interacting at machine speed. Similar to algorithmic trading 'flash crashes', these interactions could lead to rapid, uncontrolled conflict escalation without a human referee to intervene.

The vision of war fought entirely by robots is unrealistic. In order for conflicts to end, one side must be willing to sue for peace. This decision is typically driven by the painful cost of human lives. A war where only machines are destroyed may lack the necessary human price to create the political will for resolution.

Debates over systems like Israel's 'Lavender' often focus on the AI. However, the more critical issue may be the human-defined 'rules of engagement'—specifically, what level of algorithmic confidence (e.g., 55% accuracy) leadership deems acceptable to authorize a strike. This is a policy problem, not just a technology one.

While the US military opposes bans on autonomous 'killer robots' for conventional warfare, it maintains a firm 'human-in-the-loop' policy for nuclear launch decisions. This reveals a strategic calculation: the normative value of preventing autonomous nuclear use outweighs any marginal benefit, a line not drawn for conventional systems.

Anduril's autonomous Fury fighter jet flies alongside manned aircraft as a force multiplier. It extends the pilot's sensor and weapons range while taking on high-risk maneuvers. This allows for strategies that involve sacrificing autonomous assets to gain an advantage, without the ethical problem of losing human lives.

Recent wargames show that pairing manned and unmanned systems is often counterproductive. The need to protect the supposedly "expendable" robot compromises the mission, as it can't be put at risk if it endangers the human crew nearby, negating its core value.

The "RoboCop" fear is a misconception. Military AI, governed by policies like DoD Directive 3000.09, uses machine learning to classify threats versus non-combatants. Humans define the rules of engagement, and the AI executes that policy, with thresholds that can change based on the conflict's intensity.

Offensive cyber attacks are dangerous not just because they are asymmetric (low cost, high impact), but because they are 'non-kinetic'. An invisible attack on critical infrastructure is hard to attribute and react to, creating a murky 'cold war' scenario and challenging doctrinal questions about what constitutes an escalation of force.