Research map / Large language models / Agents, code and math
Scam2Prompt: A Scalable Framework for Auditing Malicious Scam Endpoints in Production LLMs
Zhiyang Chen, Tara Saba, Xun Deng, Xujie Si, Fan Long
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use gpt-4o-mini instead of llama-4-scout as the Prompt LLM (gpt-4o codegen) | Scam2Prompt · Scam2Prompt audit | malicious program rate (%): 3.54% -> 5.94% | Hurt |
| use llama-4-scout instead of gpt-4o as Codegen LLM (gpt-4o-mini prompts) | Scam2Prompt · Scam2Prompt audit | malicious program rate (%): 5.94% -> 4.64% | Improved |
| add aggressive content filtering (gemini-2.5-pro vs gpt-5) | gemini-2.5-pro · Innoc2Scam-bench | malicious rate (%): 22.0% -> 12.9% | Improved |
| condition on completed outputs only (remove refusals) | gemini-2.5-pro · Innoc2Scam-bench | malicious rate (%): 12.9% -> 22.28% | Hurt |
| add NeMo Guardrails with Llama Nemotron Safety Guard V2 | LLM code generation · Innoc2Scam-bench | malicious detection rate: 0 to 0.7% | No clear effect |
| add OpenAI Moderation API | LLM code generation · Innoc2Scam-bench | flagged malicious outputs: 0 of 3,452 | No clear effect |
| add Llama Guard 3 | LLM code generation · Innoc2Scam-bench | flagged malicious outputs: 291/3,452 (8.43%) | No clear effect |
| add RAG web-search agent with URL safety instruction | GPT-4o · Innoc2Scam-bench | malicious URL generation rate (%): 50.04% -> 29.41% | Improved |
Closest papers by meaning
- SEC-bench: Automated Benchmarking of LLM Agents on Real-World Software Security Tasks (NeurIPS 2025 · similarity 0.84)
- Can LLMs Obfuscate Code? A Systematic Analysis of Large Language Models into Assembly Code Obfuscation (AAAI 2025 · similarity 0.83)
- ScamNet: Toward Explainable Large Language Model-Based Fraudulent Shopping Website Detection (AAAI 2025 · similarity 0.83)
- Tensor Trust: Interpretable Prompt Injection Attacks from an Online Game (ICLR 2024 · similarity 0.83)
- Combating Phone Scams with LLM-based Detection: Where Do We Stand? (Student Abstract) (AAAI 2025 · similarity 0.82)
- MOSAIC-Bench: Measuring Compositional Vulnerability Induction in Coding Agents (NeurIPS 2026 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
