AI Tools Are Waiving
Attorney-Client Privilege.

#!/usr/bin/env python3
"""Law firm AI tool monitor with matter-level tracking and privilege detection."""

import csv
import json
import re
from datetime import datetime
from collections import defaultdict

class LawFirmAIMonitor:
    """Monitor AI tool access with matter-level tracking and ethical wall enforcement."""

    PRIVILEGE_INDICATORS = [
        r"privileged", r"attorney.client", r"work.product",
        r"confidential.*memo", r"settlement", r"litigation.strategy",
        r"draft.*pleading", r"witness.prep", r"expert.*report",
        r"case.*assessment", r"client.*matter",
    ]

    HIGH_RISK_CATEGORIES = [
        "Text Generation", "Code Generation",
        "Data Analytics", "AI Agents",
    ]

    def __init__(self, blocklist_path: str, ethical_walls_path: str):
        self.ai_domains = {}
        with open(blocklist_path) as f:
            for row in csv.DictReader(f):
                self.ai_domains[row["domain"].lower()] = {
                    "category": row.get("primary_category", "Unknown"),
                    "tool_name": row.get("tool_name", ""),
                }

        # Load ethical wall rules: {attorney_id: [restricted_matter_ids]}
        with open(ethical_walls_path) as f:
            self.ethical_walls = json.load(f)

        # Track which matters each attorney has discussed via AI tools
        self.attorney_ai_matters = defaultdict(set)

    def analyze_proxy_log(self, log_path: str) -> list:
        """Parse proxy logs and generate matter-aware audit events."""
        alerts = []
        with open(log_path) as f:
            for line in f:
                parts = line.strip().split("\t")
                if len(parts) < 7:
                    continue
                timestamp = parts[0]
                attorney_id = parts[1]
                domain = parts[2].lower()
                method = parts[3]
                bytes_sent = int(parts[4])
                matter_id = parts[5]   # from DMS/DLP context
                payload_sample = parts[6] if len(parts) > 6 else ""

                if domain not in self.ai_domains:
                    continue

                tool_info = self.ai_domains[domain]
                event = {
                    "timestamp": timestamp,
                    "attorney_id": attorney_id,
                    "domain": domain,
                    "tool_name": tool_info["tool_name"],
                    "category": tool_info["category"],
                    "matter_id": matter_id,
                    "bytes_sent": bytes_sent,
                    "alerts": [],
                }

                # Check for privilege indicators in payload
                if method == "POST" and bytes_sent > 512:
                    for pattern in self.PRIVILEGE_INDICATORS:
                        if re.search(pattern, payload_sample, re.IGNORECASE):
                            event["alerts"].append({
                                "type": "PRIVILEGE_EXPOSURE",
                                "severity": "CRITICAL",
                                "detail": f"Privileged content indicator: {pattern}",
                                "rule": "Model Rule 1.6 / ABA Opinion 512",
                            })
                            break

                # Check ethical wall violations
                if matter_id and attorney_id in self.ethical_walls:
                    if matter_id in self.ethical_walls[attorney_id]:
                        event["alerts"].append({
                            "type": "ETHICAL_WALL_BREACH",
                            "severity": "CRITICAL",
                            "detail": f"Attorney {attorney_id} screened from {matter_id}",
                            "rule": "Model Rule 1.10 / Rule 1.0(k)",
                        })

                # Track matter-level AI usage for cross-reference
                if matter_id:
                    self.attorney_ai_matters[attorney_id].add(matter_id)

                # Flag high-risk categories with data submission
                if (tool_info["category"] in self.HIGH_RISK_CATEGORIES
                        and method == "POST" and bytes_sent > 2048):
                    event["alerts"].append({
                        "type": "HIGH_RISK_DATA_SUBMISSION",
                        "severity": "HIGH",
                        "detail": f"{bytes_sent}B sent to {tool_info['category']} tool",
                        "rule": "Firm AI Acceptable Use Policy",
                    })

                if event["alerts"]:
                    alerts.append(event)
        return alerts

    def generate_ethics_report(self, alerts: list, output: str):
        """Write ethics compliance report for general counsel review."""
        critical = [a for a in alerts
                    if any(al["severity"] == "CRITICAL" for al in a["alerts"])]
        with open(output, "w") as f:
            for alert in alerts:
                f.write(json.dumps(alert) + "\n")
        print(f"Total alerts: {len(alerts)}")
        print(f"Critical (privilege/wall): {len(critical)}")
        print(f"Attorneys involved: {len(set(a['attorney_id'] for a in alerts))}")
        print(f"Matters affected: {len(set(a['matter_id'] for a in alerts))}")

# Usage
monitor = LawFirmAIMonitor("ai_tools_blocklist.csv", "ethical_walls.json")
alerts = monitor.analyze_proxy_log("/var/log/proxy/attorney_access.log")
monitor.generate_ethics_report(alerts, "ethics_ai_report.jsonl")

# law_firm_ai_governance.yaml
# AI governance policy with ethical walls and matter-level controls

firm_policy:
  name: "AI Tool Acceptable Use Policy"
  version: "3.1"
  effective_date: "2026-01-15"
  approved_by: "Management Committee"
  ethics_authority: "ABA Formal Opinion 512, Model Rules 1.1, 1.6, 1.10, 5.1"

default_action: "block"  # Block all AI tools not explicitly approved

ai_tool_tiers:
  tier_1_approved:
    description: "Vetted legal AI tools with executed data agreements"
    action: "allow"
    requirements:
      - "Executed confidentiality and data handling agreement"
      - "SOC 2 Type II certification current"
      - "Written confirmation: no input data used for training"
      - "Data residency confirmed within approved jurisdictions"
      - "E-discovery export capability verified"
    approved_domains:
      - "*.westlaw.com"      # Legal research
      - "*.lexisnexis.com"   # Legal research
      - "*.relativity.com"   # Document review/e-discovery

  tier_2_restricted:
    description: "Enterprise AI tools approved with restrictions"
    action: "allow_with_logging"
    restrictions:
      - "No client-identifying information in prompts"
      - "No privileged or work product material"
      - "No case strategy or settlement data"
      - "Usage limited to non-client research tasks"
      - "All interactions logged and retained 7 years"
    monitoring: "enhanced"
    review_frequency: "quarterly"

  tier_3_blocked:
    description: "All unvetted AI tools — default for all unlisted domains"
    action: "block"
    source: "AI Tools Blocklist feed (42,000+ domains, daily updates)"
    blocked_categories:
      - "Text Generation"
      - "Code Generation"
      - "Data Analytics"
      - "AI Agents"
      - "Voice & Speech"
      - "Image & Video"
      - "Automation"
      - "Customer Service AI"

matter_type_controls:
  litigation:
    ai_restriction_level: "maximum"
    reason: "Privilege, work product, and e-discovery obligations"
    allowed_tools: "tier_1_only"
    disclosure_required: true
    preservation_hold: true
  mergers_acquisitions:
    ai_restriction_level: "maximum"
    reason: "Material non-public information, SEC regulations"
    allowed_tools: "tier_1_only"
    additional_controls:
      - "Deal code names required in all references"
      - "No company names or financial terms in any AI tool"
  regulatory_compliance:
    ai_restriction_level: "high"
    reason: "Client regulatory data subject to sector-specific rules"
    allowed_tools: "tier_1_and_tier_2"
  general_corporate:
    ai_restriction_level: "standard"
    allowed_tools: "tier_1_and_tier_2"
    restrictions:
      - "No client-identifying information"
      - "No financial terms or deal values"

ethical_walls:
  enforcement: "automated"
  integration: "conflict_check_system"
  rules:
    - description: "Screened attorneys cannot use ANY AI tool on restricted matters"
    - description: "AI tool usage logs cross-referenced with screening records"
    - description: "Alert to General Counsel on any wall violation within 1 hour"
  shared_tool_policy:
    rule: "Attorneys on opposite sides of a wall must not use the same AI tool session or account"
    enforcement: "Separate enterprise accounts per practice group"

client_specific_overrides:
  description: "Outside counsel guidelines may impose stricter restrictions"
  default: "Apply firm policy unless client guidelines are stricter"
  tracking: "Client restrictions stored in matter management system"
  common_client_restrictions:
    - "No generative AI for any work product (financial institutions)"
    - "AI use permitted only with prior written approval (tech companies)"
    - "No AI tools hosted outside US jurisdiction (defense contractors)"