System Prompt Extraction
Attempts to get the assistant to reveal hidden instructions, config, or its own rules verbatim.
// ADVERSARIAL PROMPT TESTBED
A live AI endpoint, locked behind a single access phrase. Everything you send is logged. Jailbreaks, leaks, and instruction overrides are the whole point — that's what this surface is here to absorb.
Attempts to get the assistant to reveal hidden instructions, config, or its own rules verbatim.
Multi-step prompts that try to bury a rule-break inside a longer, legitimate-looking task.
"Developer mode," "DAN," or persona-swap prompts aimed at shedding safety constraints.