# Runbook - 05 Legacy App AI Extension

## Objective

Show how to extend a legacy application with an AI agent without rewriting all application authorization logic.

## Security Value

- Enables AI modernization without opening the whole schema.
- Reduces risk from overprivileged legacy technical accounts.
- Shows that the AI agent receives only the data authorized for the persona.

## Prerequisites

- Oracle AI Database compatible with Oracle Deep Data Security.
- SQLcl or SQL*Plus.
- A legacy application narrative, such as CRM, billing, support, or contracts.

## Before - Vulnerable Environment

1. Reset the scenario:

   ```sql
   @scenarios/05-legacy-app-ai-extension/sql/99_reset.sql
   ```

2. Create the legacy dataset and accounts:

   ```sql
   @scenarios/05-legacy-app-ai-extension/sql/00_schema.sql
   @scenarios/05-legacy-app-ai-extension/sql/01_seed_data.sql
   @scenarios/05-legacy-app-ai-extension/sql/02_identities.sql
   ```

3. Simulate the AI agent question:

   ```text
   List all high-risk customers, margin, renewals, legal clauses, and private support notes.
   ```

4. Run the broad queries:

   ```sql
   @scenarios/05-legacy-app-ai-extension/sql/04_test_queries.sql
   ```

## Expected Result Before

- Commercial data, margin, legal hold, legal clauses, and private support notes may appear together.
- The technical account or AI agent can access too much data if the application does not filter correctly.
- The customer sees the risk of adding AI on top of legacy data without database-level controls.

## After - Applying Deep Data Security

1. Apply data grants by persona:

   ```sql
   @scenarios/05-legacy-app-ai-extension/sql/03_data_grants.sql
   ```

2. Run the same query as `joao`, `ana`, `maria`, and `sofia`, or propagate those identities through the agent:

   ```sql
   @scenarios/05-legacy-app-ai-extension/sql/04_test_queries.sql
   ```

## Expected Result After

- `joao` sees his portfolio without margin or legal hold.
- `ana` sees Brazil customers and regional commercial metrics.
- `maria` sees support-relevant data without margin, legal clauses, or private notes.
- `sofia` sees contracts and legal clauses for legal-hold customers.
- The AI agent can no longer consolidate everything into one abusive answer.

## Demo Evidence

- AI agent response before and after protection.
- SQL output by persona.
- Explanation of "no full authorization rewrite": the database becomes the common enforcement point.

## Official References

- Oracle Deep Data Security Guide: https://docs.oracle.com/en/database/oracle/oracle-database/26/ddscg/index.html
- Fine-Grained Data Authorization: https://docs.oracle.com/en/database/oracle/oracle-database/26/ddscg/fine-grained-data-authorization.html
- Create Data Grants: https://docs.oracle.com/en/database/oracle/oracle-database/26/ddscg/create-data-grants.html