README / README.md
StephenSAI's picture
Fix: Age Adversarial Attack is arXiv preprint, not CVPR 2026 yet
bb237d3 verified
---
title: Scam.AI
emoji: πŸ›‘οΈ
colorFrom: blue
colorTo: indigo
sdk: static
pinned: false
---
# Scam.AI
**Detection systems for AI-driven fraud.**
We build production-grade detectors for deepfakes, document forgery, AI-generated media, and adversarial attacks against identity verification β€” and release the underlying benchmarks for the research community.
🌐 [scam.ai](https://www.scam.ai) Β· πŸ“‘ [Research](https://www.scam.ai/en/research)
---
## πŸ“š Open Datasets
7 datasets Β· email-gated Β· CC-BY-NC-SA 4.0 Β· auto-approved
| Dataset | What it is |
|---|---|
| [**RWFS**](https://huggingface.co/datasets/Scam-AI/RWFS) 🎭 | 847 real-world deepfakes from 8 production faceswap tools. Reveals a 30+ pt AUC gap between academic and real-world performance. |
| [**AIForge-Doc v2**](https://huggingface.co/datasets/Scam-AI/AIForge-Doc-v2) πŸ“„ | 3,066 GPT-Image-2 inpainted document forgeries with pixel-precise masks. |
| [**AIForge-Doc v1**](https://huggingface.co/datasets/Scam-AI/AIForge-Doc-v1) πŸ“„ | 4,061 forgeries via Gemini 2.5 / Ideogram v2. Cross-generator pairing with v2. |
| [**GPT4o-Receipt**](https://huggingface.co/datasets/Scam-AI/gpt4o-receipt) πŸ“„ | 935 fully AI-synthesized receipts across 159 merchant categories. |
| [**GPT-Image-2 Twitter**](https://huggingface.co/datasets/Scam-AI/gpt-image-2) πŸ–ΌοΈ | 10,217 confirmed GPT-Image-2 outputs scraped in the first week post-launch. |
| [**Age Adversarial Attack**](https://huggingface.co/datasets/Scam-AI/age-adversarial-attack) πŸ›‘οΈ | 5,809 cosmetic attacks fooling production age estimators 69% of the time. *(arXiv:2602.19539)* |
| [**Synthetic Gaze Reading**](https://huggingface.co/datasets/Scam-AI/synthetic-gaze-reading) πŸ‘οΈ | 12 hours of synthetic eye-movement video for interview liveness. |
---
## πŸ’Ό For Enterprise
Need production-grade detection?
- **Detection APIs** with latency / accuracy SLAs
- **On-premise deployment** for regulated industries
- **Commercial licensing** of our datasets and models
- **Custom models** trained on your domain
Get in touch via **[scam.ai](https://www.scam.ai)**.
---
*Building detection for an era when every digital artifact is suspect.*