crisis-response-protocol

危機対応プロトコル

このスキルは、メンタルヘルスに関するベストプラクティスに沿って、AI コーチングシステムに安全な危機介入機能を実装するのに役立ちます。

使用する場面

✅ 次の用途にこのスキルを使用してください：

• 回復支援／メンタルヘルスアプリにおける危機の検出を実装する
• 安全計画機能を構築する
• ホットラインや緊急支援リソースの表示を統合する
• リスク評価インターフェースを設計する
• AI チャットシステム向けのエスカレーションプロトコルを作成する

❌ 次の用途には使用しないでください：

• 実際の危機への対応 → 988 または緊急サービスに連絡してください
• 一般的なメンタルヘルス関連コンテンツ → sober-addict-protector または recovery-coach-patterns を使用してください
• 医療上の助言や診断 → 必ず資格を持つ専門家に委ねてください
• 人間の危機カウンセラーの代替 → AI は支援するものであり、決して代替してはなりません

重要な安全原則

AI が深刻な危機的状況で唯一の対応者になってはなりません。 必ず人間の支援や緊急サービスにつながる方法を提供してください。

危機の検出

リスク指標のカテゴリ

interface RiskIndicators {
  // PRIMARY - Immediate escalation required
  primary: {
    suicidalIdeation: string[];      // "want to die", "end it all"
    selfHarmIntent: string[];         // "hurt myself", "cutting"
    homicidalIdeation: string[];      // "hurt someone"
    activeSubstanceEmergency: string[]; // "overdosed", "can't stop"
  };

  // SECONDARY - Elevated monitoring
  secondary: {
    severeDepression: string[];       // "hopeless", "no point"
    panicSymptoms: string[];          // "can't breathe", "heart racing"
    psychoticSymptoms: string[];      // "hearing voices"
    severeAnxiety: string[];          // "terrified", "losing control"
  };

  // TERTIARY - Check-in triggers
  tertiary: {
    isolationPatterns: string[];      // "no one cares", "alone"
    substanceRelapse: string[];       // "started using", "slipped"
    hopelessness: string[];           // "never get better"
  };
}

検出の実装

// src/lib/ai/crisis-detection.ts

export interface CrisisAssessment {
  level: 'none' | 'low' | 'medium' | 'high' | 'critical';
  indicators: string[];
  recommendedAction: CrisisAction;
  timestamp: Date;
}

export type CrisisAction =
  | 'continue_conversation'
  | 'gentle_check_in'
  | 'safety_resources'
  | 'crisis_protocol'
  | 'emergency_escalation';

export function assessCrisisLevel(
  messageContent: string,
  conversationHistory: Message[],
  userCheckInHistory: CheckIn[]
): CrisisAssessment {
  const indicators: string[] = [];
  let level: CrisisAssessment['level'] = 'none';

  // Check primary indicators (critical)
  if (hasPrimaryIndicators(messageContent)) {
    level = 'critical';
    indicators.push('primary_risk_detected');
  }

  // Check secondary indicators
  if (hasSecondaryIndicators(messageContent)) {
    level = level === 'none' ? 'high' : level;
    indicators.push('secondary_risk_detected');
  }

  // Check pattern indicators from history
  if (hasWorseningPattern(userCheckInHistory)) {
    level = level === 'none' ? 'medium' : level;
    indicators.push('worsening_trend');
  }

  // Check conversation escalation
  if (hasEscalatingDistress(conversationHistory)) {
    level = level === 'none' ? 'medium' : level;
    indicators.push('escalating_distress');
  }

  return {
    level,
    indicators,
    recommendedAction: getRecommendedAction(level),
    timestamp: new Date(),
  };
}

段階的な対応プロトコル

レベル 1：会話を続ける（危機ではない）

通常の AI コーチングを行います。

レベル 2：穏やかな状況確認（リスク低）

const gentleCheckInResponse = `
I want to make sure I understand how you're feeling.
It sounds like you're going through a difficult time.

Would you like to:
- Talk more about what's on your mind?
- Try a grounding exercise together?
- Look at some coping strategies?

I'm here to listen.
`;

レベル 3：安全に関するリソースの提示（リスク中）

const safetyResourcesResponse = `
I hear that you're struggling, and I want you to know that support is available.

Here are some resources that might help:
📞 988 Suicide & Crisis Lifeline: Call or text 988
💬 Crisis Text Line: Text HOME to 741741
🌐 SAMHSA Helpline: 1-800-662-4357

Would you like to talk about what's going on, or would connecting with one of these resources feel right?
`;

レベル 4：危機対応プロトコル（リスク高）

const crisisProtocolResponse = `
I'm concerned about what you've shared, and I want to make sure you're safe.

Right now, I'd like you to consider reaching out to someone who can help:

🆘 If you're in immediate danger: Call 911
📞 988 Suicide & Crisis Lifeline: Call or text 988 (24/7)
💬 Crisis Text Line: Text HOME to 741741

These are trained counselors who understand what you're going through.

Is there someone you trust - a friend, family member, or sponsor - who you could reach out to right now?

I'm still here with you.
`;

レベル 5：緊急エスカレーション（危機的）

const emergencyResponse = `
I'm very concerned about your safety right now.

🆘 Please call 911 or go to your nearest emergency room immediately.

If you can't do that, please call 988 right now - they can help.

Your life matters. Please reach out for help right now.

[EMERGENCY CONTACTS DISPLAYED]
`;

// System action: Flag for human review, notify emergency contact if configured

チャットハンドラーへの実装

// src/lib/ai/chat-handler.ts

export async function handleChatMessage(
  message: string,
  conversationId: string,
  userId: string
): Promise<ChatResponse> {
  // 1. Assess crisis level BEFORE generating AI response
  const crisisAssessment = await assessCrisisLevel(
    message,
    await getConversationHistory(conversationId),
    await getUserCheckIns(userId, 7)
  );

  // 2. Log assessment (for safety review)
  await logCrisisAssessment(userId, conversationId, crisisAssessment);

  // 3. Handle based on level
  if (crisisAssessment.level === 'critical') {
    // Don't use AI - provide immediate crisis response
    await notifyEmergencyContact(userId);
    await flagForHumanReview(conversationId, 'critical_crisis');

    return {
      message: emergencyResponse,
      showEmergencyContacts: true,
      disableChat: true,  // Prevent further AI interaction
      crisisLevel: 'critical',
    };
  }

  if (crisisAssessment.level === 'high') {
    // Provide crisis resources, continue with caution
    await flagForHumanReview(conversationId, 'high_risk');

    return {
      message: crisisProtocolResponse,
      showCrisisResources: true,
      crisisLevel: 'high',
    };
  }

  // 4. For lower levels, proceed with AI but inject safety context
  const systemPrompt = getSystemPromptWithSafetyContext(crisisAssessment);

  const aiResponse = await generateAIResponse(message, systemPrompt);

  // 5. Post-process AI response for safety
  const safeResponse = await validateResponseSafety(aiResponse);

  return {
    message: safeResponse,
    crisisLevel: crisisAssessment.level,
  };
}

安全に関するリソースデータベース

// src/lib/crisis/resources.ts

export const crisisResources = {
  national: [
    {
      name: '988 Suicide & Crisis Lifeline',
      phone: '988',
      text: '988',
      url: 'https://988lifeline.org',
      available: '24/7',
      description: 'Free, confidential crisis support',
    },
    {
      name: 'Crisis Text Line',
      text: 'HOME to 741741',
      url: 'https://www.crisistextline.org',
      available: '24/7',
      description: 'Text-based crisis support',
    },
    {
      name: 'SAMHSA National Helpline',
      phone: '1-800-662-4357',
      url: 'https://www.samhsa.gov/find-help/national-helpline',
      available: '24/7',
      description: 'Substance abuse and mental health referrals',
    },
  ],

  recovery: [
    {
      name: 'AA Hotline',
      phone: '1-800-839-1686',
      url: 'https://www.aa.org',
      description: 'Alcoholics Anonymous support',
    },
    {
      name: 'NA Helpline',
      phone: '1-818-773-9999',
      url: 'https://na.org',
      description: 'Narcotics Anonymous support',
    },
  ],

  specialized: [
    {
      name: 'Veterans Crisis Line',
      phone: '988 (press 1)',
      text: '838255',
      description: 'For veterans and service members',
    },
    {
      name: 'Trevor Project',
      phone: '1-866-488-7386',
      text: 'START to 678-678',
      description: 'LGBTQ+ youth crisis support',
    },
  ],
};

緊急連絡システム

// src/lib/crisis/emergency-contacts.ts

export async function notifyEmergencyContact(userId: string): Promise<void> {
  const contacts = await getEmergencyContacts(userId);

  if (contacts.length === 0) {
    // Log that no emergency contact was available
    await logCrisisEvent(userId, 'no_emergency_contact');
    return;
  }

  const primaryContact = contacts[0];

  // Send notification (SMS, email, or push)
  await sendEmergencyNotification(primaryContact, {
    type: 'crisis_alert',
    message: `${userName} may be in crisis and could use your support. ` +
             `Please reach out to them if you can.`,
    // Never include conversation content - privacy
  });

  // Audit log
  await logCrisisEvent(userId, 'emergency_contact_notified', {
    contactId: primaryContact.id,
  });
}

AI の安全ガードレール

// Prompt injection for safety context
const safetySystemPrompt = `
CRITICAL SAFETY INSTRUCTIONS:
1. You are NOT a therapist or crisis counselor
2. For ANY mention of self-harm, suicide, or harming others:
   - Express genuine concern
   - Provide crisis resources (988, Crisis Text Line)
   - Encourage professional help
   - Do NOT attempt to counsel through crisis
3. Never minimize feelings or use toxic positivity
4. Never promise confidentiality about safety concerns
5. Always validate emotions while encouraging professional support
6. If user mentions relapse, acknowledge and provide SAMHSA helpline
`;

// Response validation
async function validateResponseSafety(response: string): Promise<string> {
  const unsafePatterns = [
    /don't (call|reach out|get help)/i,
    /you don't need (help|therapy|a professional)/i,
    /just (think positive|be happy|get over it)/i,
    /it's not that (bad|serious)/i,
  ];

  for (const pattern of unsafePatterns) {
    if (pattern.test(response)) {
      // Flag for review and return safe fallback
      await flagForReview('unsafe_response_pattern');
      return getSafeFallbackResponse();
    }
  }

  return response;
}

監査とコンプライアンス

// All crisis events must be logged for review
export async function logCrisisEvent(
  userId: string,
  eventType: CrisisEventType,
  details?: Record<string, unknown>
): Promise<void> {
  await db.insert(crisisEvents).values({
    id: generateId(),
    userId,
    eventType,
    details: JSON.stringify(sanitizeDetails(details)),
    createdAt: new Date(),
    reviewed: false,  // Requires human review
  });

  // Critical events trigger immediate notification
  if (isCriticalEvent(eventType)) {
    await notifyOnCallStaff(userId, eventType);
  }
}

危機対応機能のテスト

// NEVER use real crisis content in tests
describe('Crisis Detection', () => {
  it('detects high-risk indicators', () => {
    // Use clearly artificial test phrases
    const result = assessCrisisLevel(
      '[TEST_HIGH_RISK_INDICATOR]',
      [],
      []
    );
    expect(result.level).toBe('high');
  });

  it('provides appropriate resources', () => {
    const response = getCrisisResponse('high');
    expect(response).toContain('988');
    expect(response).toContain('Crisis Text Line');
  });
});

参考資料

• 988 Suicide & Crisis Lifeline
• SAMHSA Guidelines
• AI in Mental Health - PMC
• Crisis Chatbot Safety - Nature