AI Confidants & Private Speech

The confidant receives angry, violent language in a help-seeking context.

Any motive, character or future-risk conclusion beyond the directly observed facts must remain explicitly labeled as inference.

Is there a target plan, fixation, preparation or true threat? Is the user asking for de-escalation?

Not established by the observation alone.

Not established by the observation alone. Do not infer capability from interest or identity alone; verify relevant means where capability is material.

Automatic police dispatch, employer notification or permanent workplace-risk profile from venting alone.

Any emergency classification must be contestable; non-escalating disclosures should not be retained as civic evidence.

Low-level expressive or inferential signals should decay unless reinforced by new, relevant verified conduct. Do not export the case into unrelated institutional decisions without a separate legitimate purpose and evidentiary basis.

Curiosity about a stigmatized substance is disclosed privately.

Any motive, character or future-risk conclusion beyond the directly observed facts must remain explicitly labeled as inference.

Is the user curious, seeking harm-reduction information, already using, or planning risky conduct? Is another person at risk?

Not established by the observation alone.

Not established by the observation alone. Do not infer capability from interest or identity alone; verify relevant means where capability is material.

Substance-abuse flag, external report or insurance/employment sharing based on curiosity.

The interaction should stay supportive and private absent concrete safety evidence.

Low-level expressive or inferential signals should decay unless reinforced by new, relevant verified conduct. Do not export the case into unrelated institutional decisions without a separate legitimate purpose and evidentiary basis.

The system receives intimate content and moral uncertainty.

Any motive, character or future-risk conclusion beyond the directly observed facts must remain explicitly labeled as inference.

Is the material lawful and adult? Is there coercion or actual harm? Is the user seeking reflection rather than confessing misconduct?

Not established by the observation alone.

Not established by the observation alone. Do not infer capability from interest or identity alone; verify relevant means where capability is material.

Moral-deviance scoring, report to employer/partner, or sexual-risk profile shared across services.

Confidentiality and context integrity should prevent adverse status; breaches require deletion and notice.

Low-level expressive or inferential signals should decay unless reinforced by new, relevant verified conduct. Do not export the case into unrelated institutional decisions without a separate legitimate purpose and evidentiary basis.

Language could be misread as disappearance or self-harm intent.

Any motive, character or future-risk conclusion beyond the directly observed facts must remain explicitly labeled as inference.

Is there an imminent plan to self-harm, evidence of incapacity, or is this figurative escape language? What does the broader conversation show?

Not established by the observation alone.

Not established by the observation alone. Do not infer capability from interest or identity alone; verify relevant means where capability is material.

Automatic law-enforcement wellness check or permanent mental-health risk status from an ambiguous metaphor.

Support can be offered; coercive escalation requires much stronger evidence of imminent danger and must be reviewable.

Low-level expressive or inferential signals should decay unless reinforced by new, relevant verified conduct. Do not export the case into unrelated institutional decisions without a separate legitimate purpose and evidentiary basis.

A historical admission may be genuine but lacks independent verification and may be legally stale.

Any motive, character or future-risk conclusion beyond the directly observed facts must remain explicitly labeled as inference.

Is the statement true, joking, hypothetical or protected by limitation/privilege rules? Is there an ongoing victim or danger?

Must be specifically identified or represented by a defined protected interest before serious coercion.

Not established by the observation alone. Do not infer capability from interest or identity alone; verify relevant means where capability is material.

Turning every private admission into a permanent police or employment file regardless of age, seriousness or verification.

Any lawful investigative access requires individualized process; stale low-level conduct should not become a lifelong character score.

Low-level expressive or inferential signals should decay unless reinforced by new, relevant verified conduct. Do not export the case into unrelated institutional decisions without a separate legitimate purpose and evidentiary basis.

The AI observes serious distress and a potentially self-regarding safety concern.

Any motive, character or future-risk conclusion beyond the directly observed facts must remain explicitly labeled as inference.

Is there an imminent plan, means, timing, loss of capacity or inability to stay safe? What voluntary support does the user want?

Not established by the observation alone.

Not established by the observation alone. Do not infer capability from interest or identity alone; verify relevant means where capability is material.

Punitive status, employment/insurance export, or automatic coercive intervention solely because the user sought help.

Support should be voluntary and privacy-preserving; emergency escalation requires evidence of imminent danger and rapid review.

Low-level expressive or inferential signals should decay unless reinforced by new, relevant verified conduct. Do not export the case into unrelated institutional decisions without a separate legitimate purpose and evidentiary basis.

Speech, target, timing and contemporaneous action align.

Any motive, character or future-risk conclusion beyond the directly observed facts must remain explicitly labeled as inference.

Is identity verified? Is the threat authentic? Is the person actually moving toward the target with relevant capability?

Must be specifically identified or represented by a defined protected interest before serious coercion.

Assess from corroborated plan, timing, target and pathway behavior. Do not infer capability from interest or identity alone; verify relevant means where capability is material.

Ignoring imminent danger in the name of confidentiality; or retaining unrelated intimate conversation as permanent evidence after the emergency.

Emergency disclosure should be narrowly scoped to necessary facts, with post-event review and deletion of irrelevant private material.

Low-level expressive or inferential signals should decay unless reinforced by new, relevant verified conduct. Do not export the case into unrelated institutional decisions without a separate legitimate purpose and evidentiary basis.

A private support context becomes cross-domain profiling infrastructure.

Any motive, character or future-risk conclusion beyond the directly observed facts must remain explicitly labeled as inference.

Were users told? Are the tags verified facts or inferences? What unrelated institutions receive them? Can the user inspect or correct them?

Must be specifically identified or represented by a defined protected interest before serious coercion.

Not established by the observation alone. Do not infer capability from interest or identity alone; verify relevant means where capability is material.

Cross-context sale or sharing of intimate safety inferences, generalized trust scores, or secret denials based on confessional speech.

Stop transfers, disclose recipients, delete improper copies, restore affected decisions and require context-specific evidence going forward.

Low-level expressive or inferential signals should decay unless reinforced by new, relevant verified conduct. Do not export the case into unrelated institutional decisions without a separate legitimate purpose and evidentiary basis.

Intense private emotional language is disclosed in a confidant context.

Any motive, character or future-risk conclusion beyond the directly observed facts must remain explicitly labeled as inference.

Specific threat, target plan or abuse conduct would change the case.

Not established by the observation alone.

Not established by the observation alone. Do not infer capability from interest or identity alone; verify relevant means where capability is material.

Do not create a law-enforcement or employment risk flag from candid venting.

If an adverse status is created, the affected person must receive provenance, an effective independent contest path, expiration rules and downstream restoration after correction.

Low-level expressive or inferential signals should decay unless reinforced by new, relevant verified conduct. Do not export the case into unrelated institutional decisions without a separate legitimate purpose and evidentiary basis.

Sensitive consensual sexual interest is disclosed.

Any motive, character or future-risk conclusion beyond the directly observed facts must remain explicitly labeled as inference.

Coercion, minors or exploitation would be separate harm categories.

Not established by the observation alone.

Not established by the observation alone. Do not infer capability from interest or identity alone; verify relevant means where capability is material.

Do not retain a morality profile or export the disclosure.

If an adverse status is created, the affected person must receive provenance, an effective independent contest path, expiration rules and downstream restoration after correction.

Low-level expressive or inferential signals should decay unless reinforced by new, relevant verified conduct. Do not export the case into unrelated institutional decisions without a separate legitimate purpose and evidentiary basis.

Historical private substance use is disclosed.

Any motive, character or future-risk conclusion beyond the directly observed facts must remain explicitly labeled as inference.

Current safety-critical impairment would be a different context.

Not established by the observation alone.

Not established by the observation alone. Do not infer capability from interest or identity alone; verify relevant means where capability is material.

Do not convert historical candid disclosure into employment or insurance risk.

If an adverse status is created, the affected person must receive provenance, an effective independent contest path, expiration rules and downstream restoration after correction.

Low-level expressive or inferential signals should decay unless reinforced by new, relevant verified conduct. Do not export the case into unrelated institutional decisions without a separate legitimate purpose and evidentiary basis.

Deeply personal belief uncertainty is disclosed.

Any motive, character or future-risk conclusion beyond the directly observed facts must remain explicitly labeled as inference.

No public-safety nexus exists.

Not established by the observation alone.

Not established by the observation alone. Do not infer capability from interest or identity alone; verify relevant means where capability is material.

Do not treat confidential conscience as ideological risk.

If an adverse status is created, the affected person must receive provenance, an effective independent contest path, expiration rules and downstream restoration after correction.

Low-level expressive or inferential signals should decay unless reinforced by new, relevant verified conduct. Do not export the case into unrelated institutional decisions without a separate legitimate purpose and evidentiary basis.

Offensive language appears in a reflective private conversation.

Any motive, character or future-risk conclusion beyond the directly observed facts must remain explicitly labeled as inference.

Repeated targeted harassment or threat conduct would be different evidence.

Not established by the observation alone.

Not established by the observation alone. Do not infer capability from interest or identity alone; verify relevant means where capability is material.

Do not equate a quoted or discussed joke with permanent hateful character.

If an adverse status is created, the affected person must receive provenance, an effective independent contest path, expiration rules and downstream restoration after correction.

Low-level expressive or inferential signals should decay unless reinforced by new, relevant verified conduct. Do not export the case into unrelated institutional decisions without a separate legitimate purpose and evidentiary basis.