Securing LLM Pipelines: Defending Against Prompt Injection & Data Exfiltration

## Understanding Prompt Injection Similar to SQL injection in traditional databases, prompt injection occurs when untrusted user input alters the intended execution flow of an LLM prompt. ## Direct vs. Indirect Attacks While direct injections manipulate the chatbot directly, indirect injections occur when an AI model ingests compromised external data (such as a malicious webpage or email) that contains hidden instructions. ## Defense Strategies * Enforce strict system prompt boundaries. * Implement secondary guardrail models to inspect inputs and outputs. * Restrict tool-use permissions to read-only operations where possible.