AILWS and the Rome Statute: Can an Algorithm Commit a Crime of Aggression?

TL;DR

As Artificial Intelligence graduates from science fiction to the battlefield, the legal community faces a crisis: who goes to jail when a robot starts an illegal war? This paper examines the Rome Statute of the International Criminal Court (ICC) to determine how AI-Based Lethal Weapon Systems (AILWS) fit into the crime of aggression. The verdict? The algorithm is innocent by way of non-existence (legally speaking), shifting the entire weight of justice onto the shoulders of programmers and military commanders.

The Accountability Gap: Why AI Challenges Traditional Law

The core friction in modern international law is the transition from Human-in-the-loop to Human-out-of-the-loop systems. Traditional legal frameworks, such as the Tallinn Manual or the Geneva Conventions, rely on the concept of Mens Rea—the "guilty mind."

Current AILWS are evolving toward Artificial Super Intelligence (ASI), yet the law remains anchored to Natural Persons. The paper highlights that while an AI can perform the actus reus (the physical act of an attack), it lacks the cognitive and legal capacity to possess intent. This creates a dangerous void: if no human "intent" can be found in a black-box algorithm's decision to strike, does the crime of aggression simply evaporate?

Methodology: Mapping AI Autonomy to Legal Statues

The authors categorize AI into three distinct phases to assess legal thresholds:

1. Artificial Narrow Intelligence (ANI): Tasks like target recognition (Current state).
2. Artificial General Intelligence (AGI): Human-level reasoning (Theoretical).
3. Artificial Super Intelligence (ASI): Superhuman capacity (Future).

By applying Article 8 bis of the Rome Statute, the authors analyze specific acts of aggression—such as bombardment, blockades, or invasions—carried out by AI.

Figure 1: The paper delineates how increasing autonomy complicates the attribution of legal responsibility.

The Core Finding: Programmer vs. Operator Liability

The paper argues that responsibility doesn't disappear; it relocates.

• The Programmer's Risk: If an AILWS commits a war crime because it was trained on biased or erroneous datasets, the software developer may face liability.
• The Operator's Risk: If a commander deploys an autonomous swarm knowing its behavior is unpredictable in civilian zones, the "knowledge" element of Article 30 is satisfied.

The study explicitly rejects the idea of "Legal Personality" for AI in the short term. Imprisoning or "re-coding" a software for the sake of deterrence is deemed nonsensical in a court of law.

Experimental Context: Real-World Systems

The authors bridge the gap between theory and reality by citing existing semi-autonomous systems:

• CIWS and Patriot Missiles: Systems that operate with high degrees of autonomy in defensive postures.
• C-DAEM (Cannon-Delivered Area Effects Munition): A target-seeking munition that could technically fulfill the actus reus of bombardment under Article 8 bis 2(b).

Figure 2: Analysis of different attack types (UAVs, blockades) and their classification under the Rome Statute.

Critical Insight & Recommendations

The most striking takeaway is the call for "Moral Codes in Code." The authors recommend:

1. Ethical Algorithmic Constraints: Hard-coding LOAC principles (distinction, proportionality) into software.
2. Fail-Safe Mechanisms: Robust "kill-switches" to prevent accidental escalation (similar to nuclear safety protocols).
3. International Standardization: A global commission to define technical standards for AILWS, moving beyond the currently "mild" frameworks of the US, UK, and China.

Conclusion: Humanity Remains the Firewall

While technology moves toward the "black box," the law must remain transparent. This paper reminds us that as long as the Rome Statute identifies only "natural persons" as subjects, humanity remains the ultimate firewall of legal responsibility. We cannot outsource the "guilt" of war to an algorithm.