Why write an article about warfare to highlight The Hague, the city of peace and justice? Because in this city, the Asser Institute is closely monitoring whether warfare using AI complies with legal obligations under International Humanitarian Law (IHL).

I spoke with the Asser Institute about two ongoing research projects.

AI as a people-pleaser

In his new publication, researcher Dr Jonathan Kwik provides a pioneering examination of sycophantic military AI assistants. His paper not only provides a comprehensive analysis of why bootlicking behaviour in AI systems poses significant dangers on the battlefield, but also offers militaries an integrated list of solutions to mitigate risks, through both technical measures and practical means (like operator sensibilisation and adapting operating procedures). What that means? Jonathan Kwik explains:

"But what if such digital assistants start to prioritise pleasing their operators, over reporting facts? For example, by telling a military officer what they hoped to hear (that a military objective is cleared of civilians), instead of unwanted but correct information (that a family just moved in to take shelter from bombings)?"

Comparison between AI and humans

Researchers Klaudia Klonowska and Taylor Kate Woodcock of the Asser Instituut have conducted another interesting study. When compared to humans, AI is often promised to offer superior speed, accuracy, and certainty. This supposed superiority of AI sustains broader hopes for ‘overcoming one’s own imperfections in the wielding of violence’.

In their research, Klonowska and Woodcock examine the comparison between AI and humans. They focus primarily on the suggestion that AI could be ‘better’ at reducing harm to civilians and as a result may ensure greater compliance with international humanitarian law.

Summarising their extensive research was not that easy. I asked Copilot, and this is what it came up with:

Misleading assumptions in human/ai comparisons

“The idea that AI can simply replace humans in military tasks is misleading. For example, when unmanned aircraft were introduced, they didn’t just remove the need for pilots; instead, they changed how surveillance and targeting were done, creating new roles for humans. Using AI in the military requires careful thought about how responsibilities shift and whether important values are still protected. Human-AI collaboration doesn’t automatically combine the best qualities of both; it brings its own challenges.

Comparisons between humans and AI often ignore the fact that both are highly diverse. No two humans are the same, and AI systems also vary widely in design and function. AI performance can change over time due to updates, retraining, or changing conditions, just as human performance can be affected by stress or fatigue.”

Unpacking better performance

“Speed and accuracy are often cited as AI’s main advantages. While faster operations may seem beneficial, they can reduce the time available to check information and take precautions, increasing risks to civilians. Similarly, high accuracy in tests doesn’t guarantee the same results in real-world situations. For example, in the Gaza conflict, faster AI-driven targeting led to more civilian casualties.

Comparing AI and humans in military contexts often focuses on speed and accuracy, but these are tactical—not humanitarian—goals. All military actions must comply with International Humanitarian Law (IHL), which balances military necessity and civilian protection. Oversimplifying AI’s potential risks ignoring IHL’s complex, interrelated principles of distinction, precaution, and proportionality.”