Issue Details
Insecure Output Handling
llm02-2024 • Rank #2 in OWASP 2025
Explanation
Insecure output handling occurs when generated model content is directly trusted by applications, users, or downstream systems. Attackers can weaponize outputs into code injection, unsafe rendering, or misleading instructions that trigger harmful operations. The vulnerability is not only model correctness but the unsafe assumptions made by consuming components.
Remedy
Treat model output as untrusted input and enforce strict validation before rendering, storing, or executing it. Use contextual escaping, schema validation, and policy classifiers for high-risk content. For action-taking workflows, require deterministic checks and optional human approval before execution.
| Year | Rank | Category |
|---|---|---|
| 2024 | #2 | llm02-2024 - Insecure Output Handling |