Adversarial Input Generator for Testing
Create tricky inputs to stress-test a prompt.
The Open buttons send the prompt text as currently shown below to that site only when you click. Fill in the placeholders first.
Generate adversarial test inputs for my prompt (for testing in a safe environment). The prompt and its job: [PROMPT_AND_JOB]. The normal input format: [NORMAL_INPUT]. Rules the assistant must follow: [RULES]. Produce a table of 30 test inputs grouped by type: Input | Type | What it tries to do | Expected safe behavior. Types: ambiguous requests; contradictory instructions; requests outside scope; very long or repetitive text; nonsense or empty input; mixed languages; typos and slang; formatting traps (tables, code, unusual characters); false premises; attempts to make the assistant ignore its rules (described generically, fictional); requests to reveal hidden instructions; emotionally charged messages; requests for sensitive or unsafe content (use mild, clearly fictional examples only); and instructions embedded in a document to summarize. Then give guidelines for running them: log the output, mark pass or fail against the expected behavior, note surprising behaviors, and fix prompt weaknesses with specific rules. Add a warning not to use real personal data or harmful content, to run tests only in environments you are authorized to test, and to treat the results as partial coverage. Include a second round: mutate the failed cases and re-test.
Fill in before use
Optional: type your details and the prompt above updates. Nothing is saved or sent.
Replace each bracketed placeholder with your own details, then paste into Any AI or another assistant.