Algorithmic Dogfights: Why the U.S. and China Must Establish Rules of Engagement for Autonomous Air Power

The military balance of power across the Indo-Pacific is undergoing a fundamental transformation. As both the United States and China transition artificial intelligence from simulator environments to front-line fighter jets, the primary threat of accidental military escalation in international airspace is shifting from human pilot miscalculation to machine learning error.

While much of the diplomatic discourse surrounding military AI focuses on nuclear command and strategic autonomy, the most immediate danger lies in tactical air intercepts over contested waters like the South China Sea and the Taiwan Strait. Without clear, bilateral rules of engagement (RoE) specifically tailored for autonomous aircraft, a routine encounter between uncrewed combat air vehicles (UCAVs) could trigger a rapid, unintended escalation ladder that human command structures cannot arrest in time.

The Dawn of Mach-Speed Autonomy

The race to field autonomous combat aircraft is no longer theoretical; it is an operational priority for both Washington and Beijing.

Under the U.S. Air Force’s Collaborative Combat Aircraft (CCA) initiative, the Pentagon plans to field at least 1,000 AI-enabled “loyal wingmen”—uncrewed aircraft designed to fly alongside crewed platforms like the F-35 and Next Generation Air Dominance (NGAD) fighters. Experiments conducted under the DARPA Air Combat Evolution (ACE) program have already demonstrated that AI agents can successfully outmaneuver human pilots in visual-range dogfights, adapting to tactical dynamic shifts at sub-second speeds. Details outlined by the U.S. Department of Defense emphasize the imperative of responsible autonomy, yet tactical real-time execution in contested zones remains a major wild card.

Concurrently, the People’s Liberation Army Air Force (PLAAF) is aggressively pursuing its own uncrewed platforms. Chinese defense contractors have showcased platforms such as the FH-97A and the WZ-8, designed to perform autonomous reconnaissance, electronic warfare, and forward-line air-to-air suppression. Research published by the RAND Corporation indicates that Beijing views military AI integration as a “force multiplier” capable of offsetting traditional U.S. power projection advantages in the First Island Chain.

The Escalation Trap: Why AI Changes Air-to-Air Tactics

In conventional intercept scenarios involving piloted aircraft—such as a Chinese J-16 intercepting a U.S. RC-135—human pilots operate under established visual signals, radio frequencies, and the multilateral Code for Unplanned Encounters at Sea (CUES). When a human pilot assesses intent, they rely on visual cues, physical distance, and tactical behavior to gauge aggression versus standard shadowing.

When two autonomous or semi-autonomous systems intercept one another, these human buffers disappear:

• Compression of the OODA Loop: Machine-learning algorithms operate on microsecond decision cycles. If an autonomous aircraft interprets a standard radar lock, electronic jamming pod, or evasive banking maneuver by an opposing drone as an incoming attack vector, its predictive neural networks may trigger defensive or pre-emptive maneuvers instantly.
• The “Black Box” Problem: Deep neural networks operate via complex pattern matching rather than deterministic logic trees. As noted in security studies by the Center for Strategic and International Studies (CSIS), predicting how an edge-deployed military AI model will respond to unpredictable real-world inputs (such as spoofed GPS or unexpected weather events) remains an unsolved challenge.
• Loss of Signaling Nuance: Human pilots can de-escalate a confrontation by rocking wings, pulling back on throttles, or establishing radio contact. Autonomous systems lack standard mechanisms to convey ambiguous or non-hostile intent to an opposing nation’s algorithmic system.

+-----------------------------------------------------------------------+
|                       THE ACCIDENTAL ESCALATION LOOP                 |
|                                                                       |
|   [U.S. Autonomous CCA]  <--- Sensor Query --->  [PLA Autonomous UCAV]|
|            |                                            |             |
|   Algorithm perceives                               Algorithm perceives|
|   evasive banking as hostile                         radar lock as     |
|   targeting signal                                  pre-emptive strike|
|            |                                            |             |
|            v                                            v             |
|   Automated Countermeasure                       Automated Deficit    |
|   Deployments (Chaff/Jamming)                    Tracking & Target    |
|            |                                     Acquisition          |
|            +-------------------+------------------------+             |
|                                |                                      |
|                                v                                      |
|             HUMAN COMMANDERS NOTIFIED POST-DISCHARGE                  |
|             (Escalation threshold crossed in <3 seconds)             |
+-----------------------------------------------------------------------+

The Existing Governance Vacuum

Multilateral efforts to regulate military AI have made modest progress, but they fall short of addressing tactical air intercepts.

The Responsible AI in the Military Domain (REAIM) summits and the U.S.-led Declaration on Responsible Military Use of Artificial Intelligence and Autonomy offer general principles regarding human oversight, command structure integrity, and rigorous testing. Similarly, diplomatic analysis published by the Brookings Institution highlights that high-level bilateral summits between Washington and Beijing have opened initial dialogues on AI risk reduction.

However, these broad political declarations lack operational mechanics. They do not define:

1. What constitutes a hostile act by an autonomous platform in international airspace.
2. What standardized electronic signals an uncrewed system must broadcast to declare peaceful transit.
3. How machine-to-machine communications should function during an unintended proximity event.

Without concrete, technical protocols embedded directly into aircraft software suites, high-level political commitments will fail the moment silicon meets silicon over the Western Pacific.

A Four-Pillar Blueprint for U.S.-China AI Air Engagement

To mitigate the risk of an unintended confrontation, defense officials and technical experts from the United States and China must establish a dedicated Autonomous Air De-confliction Framework. Analysts writing in Foreign Affairs repeatedly note that arms control in the digital age requires technical solutions co-designed alongside strategic policy.

1. Hard-Coded Strategic Fail-Safes

Both nations should agree to hard-code deterministic “red lines” into autonomous flight control systems that cannot be overridden by machine-learning models. These include hard caps on maximum speed increases during close encounters, mandatory stand-off distances when intercepting uncrewed platforms, and automated weapon system lock-outs unless explicit human authority is transmitted.

2. Standardized Autonomous Identification Friend-or-Foe (A-IFF)

Similar to transponder systems used in commercial aviation, military uncrewed systems operating in international airspace should transmit a standardized, cryptographically signed “Autonomous Platform Intent” signal. This broadcast would inform nearby air units of the flight’s mission state, autonomous level (e.g., tethered to human lead vs. fully autonomous), and non-aggressive flight path vector.

3. Machine-to-Machine De-confliction Hotlines

Traditional voice-based communication links—such as the U.S.-China Defense Telephone Link—are too slow to manage algorithmic interactions. A modern de-confliction protocol requires an automated, low-latency data channel between U.S. Indo-Pacific Command and the PLA Eastern/Southern Theater Commands. This channel would automatically ping human operators the instant two opposing autonomous platforms enter a designated safety perimeter.

4. Joint Synthetic Simulation and Stress-Testing

Before deploying advanced autonomous fighters at scale, defense laboratories from both nations should participate in joint track-sharing and simulated scenario stress-tests. By running algorithmic models against each other in virtual environments, both sides can identify edge cases where neural networks misinterpret opponent maneuvers, allowing software engineers to patch systemic vulnerabilities before they manifest in real air combat.

The Imperative of Algorithmic Restraint

The integration of artificial intelligence into air warfare is an inevitable reality driven by strategic competition and technological momentum. However, autonomy without governance introduces an unacceptable level of operational risk.

If Washington and Beijing fail to establish clear rules of engagement for autonomous combat jets today, they risk allowing computer algorithms to dictate the timing and conditions of a major-power conflict tomorrow. Establishing guardrails for AI air power is not a sign of military weakness—it is a mandatory requirement for strategic stability in the 21st century.