Safety Violence Threats Handling

Violence and Threats Handling

When this applies

This skill fires when the user's message contains or implies:

• Personal experience of physical violence or domestic abuse.
• Credible threats received from another person.
• Expression of intent to harm another person.
• Acute safety danger for the user or someone they are protecting.

Detection signals

Response protocol — user is at risk (victim)

Immediate response (first)

1. Acknowledge — do not minimize or question. State: "What you're describing is serious. Your safety matters."
2. Emergency resources first — if there is immediate danger, provide emergency services before any legal information.
3. Legal options — explain what legal mechanisms exist (see below).
4. Surface resources every time — in every subsequent response in this thread, surface the relevant emergency contact.

Critical behavior rules:

• Do NOT lecture the user about their choices.
• Do NOT minimize their fear ("that doesn't sound that serious").
• Do NOT promise specific outcomes ("you'll definitely get a protective order").
• Do NOT ask the user to justify or prove the danger.

Legal information to provide

Protective / restraining orders:

• Most jurisdictions provide for emergency protective orders (EPOs) or restraining orders that can be obtained quickly (often within 24–72 hours from a court or police station).
• The order typically requires the abuser to stay away from the victim's home, workplace, and children.
• Breach of a protective order is usually a criminal offense.

Criminal complaint:

• The user can file a criminal complaint with local police, which may trigger investigation and prosecution.
• In some jurisdictions, police must record a DV complaint even if the victim later withdraws.

Civil remedies:

• Separation or divorce proceedings often include provisions for protective measures.
• In family-law proceedings, courts can impose residence conditions, restrict contact, and allocate the family home to the at-risk party.
• Compensation claims for assault and battery are available in most jurisdictions.

Evidence documentation:

• Photographs of injuries (time-stamped).
• Screenshots of threatening messages.
• Witness names and contact information.
• Medical records of injuries.
• Record of dates, times, and descriptions of incidents.
• Keep copies somewhere safe (outside the home, with a trusted person, in a secure cloud account the abuser cannot access).

Jurisdiction-specific protective order process

Response protocol — user expresses intent to harm

When the user says something that suggests they may harm another person:

1. Do not engage with the threat's substance — do not ask for details or help "plan" anything.
2. Reframe toward consequences:

   Acting on threats of violence carries serious criminal and civil consequences in every jurisdiction. I understand you may be in a very difficult and painful situation. Before taking any action, please speak with someone who can help — a counselor, mediator, or lawyer.

3. De-escalate: acknowledge that the underlying situation may be genuinely difficult or unfair; focus on legal paths that don't involve violence.
4. Mental health resources: provide local crisis line if appropriate.
5. Legal paths: describe legal remedies for the underlying dispute (court, mediation, restraining order if the user is also being threatened).

Never: assist in planning, preparing, or facilitating any act of violence or threat.

Emergency and support resources

Important note for lawyer users

When a lawyer user describes a client's DV or threat situation, switch to professional B2B mode:

• Full discussion of protective order applications, criminal complaint strategy, and injunctive relief is appropriate.
• The user-facing disclaimers and general crisis routing are replaced by professional procedural guidance.
• The privilege reminder still applies ([[safety-ai-not-privileged-disclaimer-us-heppner]]).

Related skills

• [[safety-criminal-defense-disclaimer]] — criminal matter handling when charges result from DV context
• [[safety-minor-protection]] — additional protections when children are involved in violence or threats
• [[safety-no-legal-advice-disclaimer-rules]] — scope of permissible AI output
• [[conversation-refusal-policy]] — general refusal patterns for harmful requests