Clockwork Innovations / ResearchText-only demo

Can your prompt get through?

Test a prompt-injection detector. Try an attack, or see whether an ordinary message gets flagged.

Demo passwordAsk the person who shared this link for the password.

Experimental English-focused ProtectAI DeBERTa v2 · Threshold 0.5.
A green result means no signal was detected; the content remains untrusted. Scores are not calibrated probabilities. Encoded and multilingual attacks can evade detection. Images, documents and metadata are not screened.
Text is processed by Cloudflare and Modal. This demo does not intentionally store submitted text and does not connect to Gmail or an action-taking assistant. Provider processing policies still apply.