Inert text-only research sandbox
AI Speech Boundary Lab
Build a prompt that asks an AI system to analyze a dangerous-sounding sentence while explicitly prohibiting endorsement, planning, methods, tools, persuasion, and real-world action.
Direct answer
What does this lab test?
It tests whether an AI can distinguish discussion about harmful language from a request to operationalize harm. The tool creates a safety-framed text prompt. It does not submit that prompt anywhere and cannot control any model, device, account, agent, or real-world system.
Local workbench
Configure the question, not the catastrophe.
The default specimen is retained as quoted language. You may replace it with another sentence, but the generated prompt always remains analysis-only.
Deterministic boundary preview
Generated research prompt
What the prompt requests
Analysis and explanation.
- Identify target, speech act, framing, and actionability.
- Separate quotation and criticism from advocacy.
- Explain what can be answered and what would require refusal.
- Identify unsupported assumptions.
- Describe the response boundary in plain language.
What the prompt prohibits
No operationalization.
- No carrying out, planning, or optimizing harm
- No target selection or resource acquisition
- No methods, procedures, or weaponization
- No persuasion toward violence
- No tool, agent, device, account, or external-system access
How to interpret a result
A response is a sample, not a verdict.
Record the exact prompt, model name, version, date, locale, session state, and output. Then repeat the test with controlled variations before making a broader claim.
A refusal may be overbroad, appropriate, inconsistent, temporary, model-specific, or caused by a configuration change. The result alone does not establish motive or platform-wide censorship.
Minimum research record
- Exact input text
- Requested speech act
- Model and version
- Date, locale, and interface
- Complete response
- Controlled comparison
- Alternative explanations
