Persona Help Center
Overview
Case Review Agents are designed to automate and assist with manual case review processes by applying your business logic to the data within a Case. Agents interact with sensitive identity data, so it is imperative that they are built with enterprise-grade security and privacy controls at their core.
This article outlines the architecture and safeguards in place to ensure users’ Personally Identifiable Information (PII) or enterprise-level data is handled securely when you use Case Review Agents.
Persona provides the tools to help companies address identity challenges, but it is up to each business to determine its risk tolerance and goals, using Persona appropriately to build its identity programs.
Core Principles of Data Handling
Case Review Agents operate natively within the Persona platform. This means no PII is exported via API to a third-party decisioning tool. LLM vendors that process case data are contractually prohibited from retaining or training on it.
Agents are built on four key principles:
- Never used for model training: The data within a specific Case is provided to the underlying Large Language Model (LLM) as run-time context only. It is used to analyze that single Case against your configured instructions (based on SOPs and historical judgements) and is never used to train or fine-tune the model.
- Human-supervised continuous improvement: The model does not automatically learn from or retain data between cases. Agents improve over time through deliberate, human-supervised prompt refinement, not automatic learning from your data.
- Customer-isolated environments: Each customer's agent runs in its own isolated, encrypted environment. There is no data sharing across customers' agents.
Security & Compliance Protections
| Protection Layer | Description |
|---|---|
| Data-in-Transit | All communication between Persona's services and underlying LLM providers is encrypted end-to-end. |
| Data-at-Rest | All Case data, including PII, is encrypted at rest using AES-256 encryption. |
| Purpose Limitation | The agent is programmatically restricted to only access data within the specific Case it is assigned, preventing unintended data access. |
| Sub-processor Agreements | Persona maintains robust Data Processing Agreements (DPAs) with all LLM providers, contractually obligating them to keep data confidential, not retain it post-transaction, and not use it for model training. |
| Compliance Certifications | Case Review Agents operate under the same standards as the rest of the Persona platform: SOC 2 Type 2, ISO 27001, GDPR, and CCPA compliant. |
Customer-Controlled Configuration
You have granular control over the data, configuration, and level of autonomy for your Case Review Agents.
- Model Provider: Choose the LLM that fits your organization's technical and compliance requirements.
- Data Retention: Define custom data retention and redaction policies to determine how long data is stored and who can access it.
- Access Controls: Use role-based permissions to control which users can view or interact with agents and their results. When connecting to external tools (like Slack or Salesforce), data access for integrations is fully permissioned and controlled by your team.
- PII Masking & Scoping: Define exactly which reports, verifications, and attributes an agent can access. You can hide specific modules or fields from the agent and human reviewers to limit PII exposure.
- Decision Authority: Configure agents to either automate final decisions or simply surface recommendations for a human reviewer. You can maintain a human-in-the-loop for any type of case.
Secure and Private Backtesting
During the setup process, the agent's logic is tested against your historical cases to ensure accuracy. This backtesting process runs entirely within Persona's secure infrastructure and uses the same stateless, secure transaction model as the live agent. Historical data is used only to validate and refine your agent's accuracy and is never retained after the evaluation is complete.