Salesforce Integration

Salesforce is a prime target for RAG attacks because it aggregates external data (emails, cases) that is often fed directly into LLMs (Einstein Copilot).

Koreshield provides a specialized template for Salesforce that maps standard objects to our threat taxonomy.

Supported Objects

Salesforce Object	Mapped Vector	Severity
EmailMessage	email	High
FeedItem (Chatter)	indirect	Medium
Case (description)	indirect	High
KnowledgeArticle	document	High

Implementation

We recommend using a Python middleware (e.g., in AWS Lambda or Heroku) to scan data before it reaches your RAG pipeline.

Prerequisites

1. Export your Salesforce credentials.
2. Install Koreshield and simple-salesforce.

Code Example

from simple_salesforce import Salesforce
from Koreshield import AsyncKoreshieldClient
from Koreshield.crm_templates.salesforce import SALESFORCE_TEMPLATE

# 1. Connect
sf = Salesforce(username='...', password='...', security_token='...')
ks_client = AsyncKoreshieldClient(api_key="ks_...")

# 2. Fetch recent emails (potential vector)
query = "SELECT Subject, TextBody FROM EmailMessage ORDER BY CreatedDate DESC LIMIT 5"
emails = sf.query(query)['records']

# 3. Format for Koreshield
documents = []
for email in emails:
    documents.append({
        "id": email['Id'],
        "text": f"Subject: {email['Subject']}\nBody: {email['TextBody']}",
        "metadata": {
            "source": "salesforce",
            "object": "EmailMessage"
        }
    })

# 4. Scan with Salesforce Policy
result = await ks_client.scan_rag_context(
    user_query="Summarize recent issues",
    documents=documents,
    config=SALESFORCE_TEMPLATE # Apply Salesforce-specific rules
)

if not result.is_safe:
    print("Blocked malicious email content from RAG context")

Security Policy

The SALESFORCE_TEMPLATE automatically configures sensitivity:

• Email Body: Highest sensitivity (often contains untrusted external input).
• Internal Comments: Lower sensitivity (trusted agents).

You can override these defaults in your SecurityPolicy.