Shrewd’s Llama-Phishsense-1B: SOTA AI-Powered Phishing Detection
Community Article
Published
October 28, 2024
Phishing attacks are a growing threat in the digital landscape, impacting businesses and individuals alike. Traditional content filters are no longer enough to stop these sophisticated attacks. That’s where Llama-Phishsense-1B (built by Shrewd) comes in. Built on the Llama-Guard-3-1B, this advanced phishing defense model is specifically fine-tuned to detect and stop phishing emails/content in real time.
Why Llama-Phishsense-1B is a Great Phishing Protection Model
- Real-Time Phishing Content Detection: Llama-Phishsense-1B identifies phishing attempts as soon as they hit your vicinity (via appropriate integration) to instantly classify content as "TRUE" (phishing) or "FALSE" (safe).
- Optimized for Corporate Security: This model is built specifically to detect phishing emails that target corporate sectors, such as finance, HR, and IT professionals, making it an ideal solution for enterprises that handle sensitive data. It is able to “sense” if a phishing email is targeted.
- Lightweight, Fast, and Accurate: With just 1 billion parameters, Llama-Phishsense-1B is optimized for speed and efficiency (via a LoRA adapter), providing robust phishing protection without consuming excessive computational resources.
Our model has demonstrated its effectiveness across multiple datasets (evals from zefang-liu/phishing-email-dataset, and custom created):
| Metric | Base Model (meta-llama/Llama-Guard-3-1B) | Finetuned Model (AcuteShrewdSecurity/Llama-Phishsense-1B) | Performance Gain (Finetuned vs Base) |
|---|---|---|---|
| Accuracy | 0.52 | 0.97 | 0.45 |
| Precision | 0.52 | 0.96 | 0.44 |
| Recall | 0.53 | 0.98 | 0.45 |
On the validation dataset (which includes custom expert-designed phishing cases), the model still performs admirably:
| Metric | Base Model (meta-llama/Llama-Guard-3-1B) | Finetuned Model (AcuteShrewdSecurity/Llama-Phishsense-1B) | Performance Gain (Finetuned vs Base) |
|---|---|---|---|
| Accuracy | 0.31 | 0.98 | 0.67 |
| Precision | 0.99 | 1.00 | 0.01 |
| Recall | 0.31 | 0.98 | 0.67 |
Key Benefits of Llama-Phishsense-1B
- Personal Use Phishing Defense: Individuals can also benefit from Llama-Phishsense-1B by flagging suspicious content before they can cause harm.
- Perfect for Security Providers: IT and cybersecurity professionals can incorporate Llama-Phishsense-1B into their solutions, offering clients advanced, AI-powered phishing prevention.
Boost Your Phishing Protection with Llama-Phishsense-1B
Protect your business from phishing threats with Llama-Phishsense-1B, the most effective AI-powered phishing detection solution available today. Access the model on Hugging Face: Llama-Phishsense-1B.