ReDence uses local models to detect social-engineering scams
Original title: ReDence: A Hybrid Agentic and Decision-Model Framework for Rapid, Evidence-Grounded Detection of Modern Social Engineering Scams
Authors report 96% accuracy across twelve scam categories and at least four times the speed of comparable agentic models.
Researchers present ReDence, an open-source hybrid framework for evidence-grounded detection of online social-engineering scams. A fine-tuned multimodal Investigator, Qwen3.5-4B, plans investigations and writes reports, while a specialized Decider, Tev1, makes binary and multiple-choice choices such as browser actions and whether to continue. The system emits structured reports of actions, supporting and contradictory evidence, and artifacts for analyst audit and takedown.
Unconfirmed: The 96% accuracy and speed figures are the authors' own evaluation results.
- The Investigator is a locally hosted fine-tuned Qwen3.5-4B model.
- Tev1 decides browser actions, evidence assessment, and whether to continue.
- Reports document actions, contradictory evidence, and collected artifacts.
- Authors say it is at least four times faster than comparable agentic models.
- Products
- ReDence
- AI models
- Qwen3.5-4BTev1
From the original article
1 min read at arxiv.orgDetecting online social engineering (SE) scams often requires examining website workflows, verifying claims against external sources, and connecting evidence across multiple pages. Existing detectors provide limited support for these investigations, leaving analysts to manually gather and validate evidence, which inevitably delays detection and subsequent takedown of these attacks.
By Jaishnoor Kaur · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.