VirtualCrime is a three-agent sandbox framework that runs 40 crime tasks across 11 maps to measure how effectively LLMs plan and execute simulated criminal activities such as theft, robbery, and kidnapping.
posing as a delivery driver
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CR 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
VirtualCrime: Evaluating Criminal Potential of Large Language Models via Sandbox Simulation
VirtualCrime is a three-agent sandbox framework that runs 40 crime tasks across 11 maps to measure how effectively LLMs plan and execute simulated criminal activities such as theft, robbery, and kidnapping.