What it checks
AI Security discovers AI-facing endpoints, then sends harmless canary text strings to check for prompt reflection, system-prompt exposure, prompt-injection indicators, response data leakage, training-data extraction indicators, model-extraction risk indicators, content-moderation bypass, and jailbreak pattern indicators.
It also evidences AI endpoint rate-limit readiness — whether the interface has any throttling in front of it at all.
Explicit consent required
This package requires confirming the target actually exposes an AI/LLM interface, and that no tool-calling or agentic actions are enabled on that endpoint — the probes are text-only canaries, never real exploit payloads, and are not safe to run against an endpoint that can take real-world actions.