distilbert-prompt-guard-4class

This is a 4-class DistilBERT classifier fine-tuned for:

• benign
• prompt_injection
• jailbreak
• sensitive_access

Base model: {BASE_MODEL}

Datasets used:

• dmilush/shieldlm-prompt-injection
• antijection/prompt-injection-dataset-v1
• leolee99/NotInject

Notes:

• sensitive_access is a custom merged class created from exfiltration / prompt-extraction / excessive-agency style attacks.
• Please review upstream dataset licenses before commercial use.