Prompt Injection Attacks: The XSS of the LLM Era

The Problem

Prompt injection is what happens when user-controlled input reaches an LLM’s context in a way that overrides or subverts the developer’s intended instructions.

Why Defense Is Hard

Unlike SQL injection — where parameterized queries cleanly separate code from data — there is no native “parameterized prompt.” The model processes instructions and data in the same channel.

Mitigations

• Input sanitization — filter known injection patterns (limited effectiveness)
• Privilege separation — LLM agents should have minimal permissions
• Output validation — validate model output before acting on it
• Monitoring — log all prompts and outputs; detect anomalies