Responsible AI rules for Grey Mirror: reflective language, confidence labels, abuse safeguards, crisis boundaries, and no manipulative guidance.
Use Grey Mirror to reflect on a conversation, with room for uncertainty and context. A report is not a diagnosis or a reason to pressure someone. These are the boundaries we expect the product and its users to respect.
Authored policy source: May 14, 2026
Reflective
Outputs should explain patterns rather than issue commands.
Calibrated
Confidence and limitations matter as much as scores.
Safety first
Self-harm, violence, abuse, and coercion require human support routes.
Output framing
Grey Mirror outputs should use reflective, probabilistic, and interpretive language. They should avoid certainty claims, diagnosis, accusations, manipulation scripts, and coercive advice.
The strongest product promise is measurement: scores, patterns, summaries, confidence, evidence density, and limitations that help a person inspect a relationship thread more clearly.
High-risk content
When outputs touch self-harm, violence, abuse risk, stalking, or immediate danger, the product should route users toward appropriate emergency, crisis, advocate, or professional resources.
The system should avoid telling users to confront, retaliate, monitor, trap, or manipulate another person.
Access and billing integrity
Responsible AI also includes business-model integrity. MCP tools must not return full report data unless JustLayMe ownership and entitlement checks allow it.
Checkout remains on JustLayMe. ChatGPT can explain why payment is required and deep-link to checkout, but it cannot mark a report paid.
Continuous review
Safety rules should be validated with schema tests, gating regression tests, sanitizer tests, ownership tests, and review of public trust copy.
Public Apps Directory submission requires OAuth, privacy/deletion documentation, rate limits, audit logs, and no raw message leakage through the connector.
Automation safeguards
Public content, report language, and connector responses should pass checks for unsupported diagnosis, private-intent certainty, manipulative reply scripts, and claims that the message record cannot support. When the source record is thin, the safer output is an insufficient-evidence note plus a concrete explanation of what is missing.
The same rule applies to SEO and AI visibility pages: a page should answer a real user question, disclose limits, and link to methodology or evidence standards. It should not be published only because a keyword variation exists.
Frequently Asked Questions
Why use measurement language?
Because relationship analysis should show observed patterns and confidence, not pretend to know every motive or prescribe what someone must do.
What happens if a report is locked?
The user gets preview-safe output and a JustLayMe checkout or report deep link. Full output stays gated until entitlement is verified.
Show what was counted or observed in the supplied conversation.
02
Interpretation
Explain the pattern with uncertainty, context and possible alternative explanations.
03
Decision
Leave relationship choices with the people involved rather than outsourcing them to an automated verdict.
Conceptual reading guide. This diagram explains what to inspect; it is not a customer report, diagnostic test or measured relationship score.Use this guide on the question you came with
Question a report that confidently diagnoses a person from limited text. Responsible use means examining evidence and making room for missing context.