AI for Legal Operations: Contract Review and Clause Extraction

{ "contract_type": "NDA | MSA | SOW | Purchase Agreement", "parties": ["company1", "company2"], "effective_date": "2024-05-15", "termination_clause": { "notice_period_days": 30, "automatic_renewal": true, "termination_for_cause": "description" }, "liability": { "liability_cap_enabled": true, "cap_amount": 100000, "cap_currency": "USD", "cap_relative_to": "annual fees" }, "indemnification": { "party_1_indemnifies": true, "party_2_indemnifies": true, "scope": "third-party IP claims" }, "confidentiality": { "nda_included": true, "return_of_info_required": true, "survival_period_days": 365 }, "deviations_from_template": [ { "deviation": "Liability cap is $500k instead of $1M", "severity": "medium", "requires_approval": true } ] }

resource "aws_dynamodb_table" "contract_extractions" { name = "contract-extractions" billing_mode = "PAY_PER_REQUEST" hash_key = "contract_id" range_key = "timestamp" attribute { name = "contract_id" type = "S" } attribute { name = "timestamp" type = "N" } }

import anthropic import json import boto3 import base64 from typing import TypedDict s3 = boto3.client("s3") dynamodb = boto3.resource("dynamodb") table = dynamodb.Table("contract-extractions") class ContractTerms(TypedDict): contract_type: str parties: list[str] effective_date: str termination_clause: dict liability: dict indemnification: dict confidentiality: dict deviations_from_template: list[dict] def extract_contract_terms(pdf_bucket: str, pdf_key: str) -> dict: """Extract key terms from a contract PDF using Claude.""" client = anthropic.Anthropic() # Download PDF from S3 response = s3.get_object(Bucket=pdf_bucket, Key=pdf_key) pdf_content = response["Body"].read() # Encode to base64 for Claude pdf_base64 = base64.standard_b64encode(pdf_content).decode("utf-8") # Call Claude with structured output message = client.messages.create( model="claude-3-5-sonnet-20241022", max_tokens=2048, messages=[ { "role": "user", "content": [ { "type": "document", "source": { "type": "base64", "media_type": "application/pdf", "data": pdf_base64 } }, { "type": "text", "text": """You are a legal contract analyst. Extract the following key terms from this contract: 1. Contract type (NDA, MSA, SOW, Purchase Agreement, etc.) 2. Parties to the contract 3. Effective date (YYYY-MM-DD format) 4. Termination clause details (notice period, auto-renewal, for-cause conditions) 5. Liability limitations (cap amount, cap relative to what, any exclusions) 6. Indemnification (who indemnifies whom, scope) 7. Confidentiality (NDA included?, return of info?, survival period) Also compare against our standard template and flag any deviations. Severity levels: - "critical": affects enforceability or our risk profile - "medium": unexpected but not immediately problematic - "low": style/formatting differences, minor changes Return valid JSON.""" } ] } ], response_format={ "type": "json_schema", "json_schema": { "name": "ContractTermsExtraction", "schema": { "type": "object", "properties": { "contract_type": {"type": "string"}, "parties": {"type": "array", "items": {"type": "string"}}, "effective_date": {"type": "string"}, "termination_clause": { "type": "object", "properties": { "notice_period_days": {"type": "integer"}, "automatic_renewal": {"type": "boolean"}, "termination_for_cause": {"type": "string"} } }, "liability": { "type": "object", "properties": { "liability_cap_enabled": {"type": "boolean"}, "cap_amount": {"type": "number"}, "cap_currency": {"type": "string"}, "cap_relative_to": {"type": "string"} } }, "indemnification": { "type": "object", "properties": { "party_1_indemnifies": {"type": "boolean"}, "party_2_indemnifies": {"type": "boolean"}, "scope": {"type": "string"} } }, "confidentiality": { "type": "object", "properties": { "nda_included": {"type": "boolean"}, "return_of_info_required": {"type": "boolean"}, "survival_period_days": {"type": "integer"} } }, "deviations_from_template": { "type": "array", "items": { "type": "object", "properties": { "deviation": {"type": "string"}, "severity": {"type": "string"}, "requires_approval": {"type": "boolean"} } } } }, "required": ["contract_type", "parties", "termination_clause"] } } } ) return json.loads(message.content[0].text) def lambda_handler(event, context): bucket = event["Records"][0]["s3"]["bucket"]["name"] key = event["Records"][0]["s3"]["object"]["key"] contract_id = key.split("/")[-1].split(".")[0] try: # Extract terms terms = extract_contract_terms(bucket, key) # Store in DynamoDB import time table.put_item( Item={ "contract_id": contract_id, "timestamp": int(time.time()), "extracted_terms": json.dumps(terms), "source_file": key, "requires_review": len([d for d in terms.get("deviations_from_template", []) if d["requires_approval"]]) > 0 } ) return { "statusCode": 200, "body": json.dumps({"contract_id": contract_id, "extracted": True}) } except Exception as e: print(f"Error processing {key}: {str(e)}") raise

# Log all approvals to CloudTrail via Lambda execution role import json def log_approval(contract_id: str, approver: str, changes: list[dict]): cloudwatch = boto3.client("logs") cloudwatch.put_log_events( logGroupName="/aws/legal-ops/contract-approvals", logStreamName=f"{contract_id}", logEvents=[ { "timestamp": int(time.time() * 1000), "message": json.dumps({ "contract_id": contract_id, "approver": approver, "approved_at": datetime.now().isoformat(), "changes": changes }) } ] )