PAWS, the Protection Assistant for Wildlife Security, is an AI system that helps park rangers fight poaching by deciding where and how to patrol. It was developed by Milind Tambe and Fei Fang, originally at the University of Southern California and continued at Harvard’s Teamcore lab, and is one of the founding applications of what Fang named “green security games.”
The system rests on two ideas. First, machine learning: PAWS analyzes years of historical patrol and poaching records together with terrain and geographic data to predict where poachers are most likely to strike next, producing risk maps. Second, game theory: because a fixed or predictable patrol schedule is easy for poachers to learn and exploit, PAWS models the rangers-versus-poachers interaction as a Stackelberg security game and generates randomized patrol routes that are efficient yet hard to anticipate. The same framework extends to illegal logging and fishing.
PAWS has moved from research prototype to real deployment. In field tests at Cambodia’s Srepok Wildlife Sanctuary beginning in late 2018, rangers following PAWS-suggested routes found over 1,000 snares in the first month - more than double the prior rate - and seized chainsaws, motorbikes, and a truck. The system has been integrated with SMART, conservation patrol software used across hundreds of protected areas, with deployments including Uganda’s Queen Elizabeth National Park and sites across Southeast Asia, supported by Microsoft’s AI for Earth program.
Why business readers should care: PAWS is a rare real-world use of game theory against an adversary that adapts. The lesson generalizes to security and fraud problems broadly - against a thinking opponent, predictability is the vulnerability, and deliberate randomization is a defense.