Detects prompt injection attacks in text using a DistilBERT model fine-tuned on 476K adversarial samples. Runs entirely in your browser.