Test our Rust-native threat detection engines in real time. This playground connects to the Spectorn gateway to evaluate prompt attacks, jailbreaks, and data leaks with low latency.
Public demo: up to 6 scans per minute. Input is processed transiently and is not retained for later explanation lookup.
No results yet
Run Guard to evaluate
The playground is a bounded product surface, not a benchmark certificate. Use it to inspect what the gateway reports, how a verdict is explained, and where policy boundaries appear. A single result is evidence about one input under one policy version; it is not proof that every attack will be detected or that every harmless prompt will pass.
The public Guard evaluates prompt-injection, jailbreak, data-exposure, and policy-risk signals. A block means the configured policy crossed its threshold. A pass means no blocking condition was returned for that request; it is not a guarantee that the content is universally safe.
Read the verdict together with the score, matched spans, engine and rule names, mode, and latency. Those fields make the decision inspectable. Context still matters, so a production policy should be tuned against representative traffic and reviewed for both false positives and missed attacks.
Use synthetic prompts and remove credentials, private tokens, personal data, customer records, and confidential system instructions. The public surface is for bounded evaluation. Sensitive material belongs only in an explicitly approved environment with its own retention, access, and incident-handling contract.
The Global playground can make a bounded anonymous scan, but that does not create an account, issue an API key, prove regional authentication, authorize production use, or open checkout. Those decisions remain behind independent live gates.