Phone scammers fight bots that learn smarter strategies over time
Game of Phones: Harnessing Game Theory and LLMs in 'Bot Wars' to Counteract Phone Scams
Cryptography and SecurityArtificial IntelligenceComputation and LanguageComputer Science and Game Theory
Summary
Phone scams cause big problems worldwide, and usual ways to stop them don’t cost scammers anything. The authors created a game-like model where automated agents, powered by large language models (LLMs), learn to adapt their strategies to waste scammers’ time and resources. They tested this with hundreds of simulated phone conversations and found that adaptive agents performed much closer to an ideal balance than fixed strategies. This approach could help develop smarter scam-baiting tools that adjust their behavior dynamically without retraining.
What this means in practice
- •For cybersecurity teams: Use adaptable AI agents to engage and waste time of phone scammers, reducing their operational capacity.
- •For call center managers: Deploy automated systems that dynamically adjust responses to scam calls, improving caller screening and reducing fraud losses.
Authors
Nardine Basta, Robin Carpentier, Benjamin Zi Hao Zhao, Dali Kaafar
Abstract
Phone scams impose substantial and growing costs on individuals and financial systems worldwide, yet existing countermeasures impose no resource cost on perpetrators. Strategic scam-baiting, in which automated agents engage scammers to deplete their operational capacity, offers a proactive alternative but requires dynamic strategy adaptation that static systems cannot provide. We introduce a game-theoretic framework enabling LLMs to learn effective counter-strategies through in-context experience without fine-tuning. The framework models scammer-baiter interactions as a two-player game with empirically grounded strategies and utility functions, using Nash equilibrium as a formal benchmark for evaluating learned behaviour. Across 300 simulated dialogues and three LLM architectures, adaptive strategy selection achieves 28.9-79.7% closer alignment to Nash equilibrium than non-adaptive baselines. The code and an annotated dataset of 300 dialogues with turn-level strategy labels and utility scores are released at https://github.com/nardinebasta/GameofPhones_/ .