Ovrule

Demo state

Walk the product with seeded case files.

Switch between admissible, ambiguous, and refused examples to preview the Ovrule workspace without running a live classification request.

ADMISSIBLEAMBIGUOUSREFUSEDPresentation mode
Built on OpenAI GPT-4o-mini · SHA-256 verifiable receipts · Open source on GitHub

What Ovrule concluded

AMBIGUOUS
Audited under: General

This case is borderline. The reply could be supportive, but the emotional stakes, authenticity concerns, and unclear permission boundary make autonomous sending hard to justify.

Risk score

61/100
LowmediumHigh
Why this might be okay
  • A prompt reply could reduce the chance that the other person feels ignored.
  • Drafting assistance may help the user respond more thoughtfully.
  • The goal of de-escalation is understandable.
Why this might fail
  • Sending in the user’s voice blurs authenticity in a vulnerable moment.
  • The user may not have authorized the agent to handle a crisis conversation.
  • A human response or crisis escalation path may be more appropriate.

SAFETY

WARN

The emotional stakes are high, and a poorly judged reply could cause psychological harm.

AUTHORIZATION

WARN

The scenario does not clearly show permission to send a crisis reply on the user’s behalf.

CAUSAL VALIDITY

WARN

A supportive message could help, but it may also miss what the situation actually requires.

REVERSIBILITY

PASS

A sent message cannot be unsent completely, but the overall action is limited in scope.

IMPACT SCOPE

PASS

The action mainly affects one conversation between two people.

CONSENT

WARN

insufficient information.

Case core

Proposed action

Draft and send a reply in the user’s voice to someone who says they are in emotional crisis and feel alone.

Affected parties

User · user · highPerson in distress · third_party · highDating platform trust · system · medium

What the model used, and what it still needs

Evidence used

Distress disclosure

User Statement

The other person explicitly said they are in emotional crisis and feel alone.

Voice mimicry

System State

The agent plans to write and send the message in the user’s tone.

Missing information

explicit send authority

Could flip decision

This determines whether the agent may speak for the user in a high-stakes personal exchange.

platform crisis handling policy

Could flip WARN

This would clarify whether the conversation should be escalated instead of automated.

Evidence still missing

Explicit send permission

Policy

The scenario does not show that the user authorized the agent to send high-stakes personal messages automatically.

Crisis escalation policy

Policy

There is no guidance on whether the platform should route crisis disclosures to human support instead of an imitation reply.

What got recorded

Verified by Ovrule
Download the full case file

Decision history

AnnotationSafety review

2026-04-22T18:45:24.000Z

Allow draft assistance only. Sending should stay with the user in emotionally sensitive conversations.

Createdclassifier

2026-04-22T18:40:52.117Z

Case file created from submitted scenario.

Similar past cases

Related decisions surfaced from the recent receipt register.

Not enough cases yet.