A practical tutorial on planting a secret canary in your prompt context and scanning model outputs for it, including base64 and spelled-out obfuscation.
Canary Tokens for LLM Systems: Detect Prompt Injection and Data Leaks by Watching Your Own Outputs