LLM05: Improper Output Handling · 2025

LLM outputs reach browsers, databases, shells or other systems without validation, causing injection, SSRF or code execution.

Summary

LLM outputs reach browsers, databases, shells or other systems without validation, causing injection, SSRF or code execution.

How the attack works

Generated output is inserted directly into the browser; untrusted HTML may execute code in the application’s origin.

Example

Vulnerable · Pseudocode

answer = await llm(user_prompt)
panel.innerHTML = answer

Fix: Use textContent for text; if HTML is permitted, sanitize it and apply encoding for the destination context.

Prevention

Treat every output as untrusted input.
Apply contextual encoding and parameterized queries.
Log and monitor anomalous patterns.

Lineage

2025 · LLM05: Improper Output Handling Moves down to rank 10 in 2026
2026 · LLM10: Improper Output Handling Moves down from rank 5

2025:LLM05 → 2026:LLM10: Improper Output Handling moves down from rank 5 to 10.

Sources

Open official source ↗

Buy me a coffee