# Anticipate, Simulate, Reason (ASR): A Comprehensive Generative AI Framework for Combating Messaging Scams

arXiv: https://arxiv.org/abs/2507.17543 · HTML: https://mlsanity.com/paper/2507.17543/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Placed in Reasoning for agents, planning and code, one of 17 directions in LLM reasoning.
- Builds on 3 of its 120 references that are in the map.
- First posted on arXiv 23 Jul 2025.

## Placed in

- LLM reasoning / [Reasoning for agents, planning and code](https://mlsanity.com/lr/reasoning-for-agents-planning-and-code/)

## Builds on (most cited first, 3 of 3)

- [DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning](https://mlsanity.com/paper/2501.12948/) (2025)
- [Towards LLM-based Fact Verification on News Claims with a Hierarchical Step-by-Step Prompting Method](https://mlsanity.com/paper/2310.00305/) (2023)
- [Anatomy of an AI-powered malicious social botnet](https://mlsanity.com/paper/2307.16336/) (2023)

## Built on by (most cited first, 1 of 1)

- [ORACLE: Anticipating Scams from Partial Trajectories in Streaming App Usage](https://mlsanity.com/paper/2605.16363/) (2026)

## Abstract (arXiv)

The rapid growth of messaging scams creates an escalating challenge for user security and financial safety. In this paper, we present the textitAnticipate, Simulate, Reason (ASR) generative AI framework to enable users to proactively identify and comprehend scams within instant messaging platforms. Using large language models, ASR predicts scammer responses and delivers real-time, interpretable support to end-users. We also develop ScamGPT-J, a domain-specific language model fine-tuned on a new, high-quality dataset of scam conversations covering multiple scam types. Thorough experimental evaluation shows that the ASR framework substantially enhances scam detection, particularly in challenging contexts such as job scams, and uncovers important demographic patterns in user vulnerability and perceptions of AI-generated assistance. Our findings reveal a contradiction where those most at risk are often least receptive to AI support, emphasizing the importance of user-centered design in AI-driven fraud prevention. This work advances both the practical and theoretical foundations for interpretable and human-centered AI systems in combating evolving digital threats.