OWASP's field guide to LLM security risks

What it does

This is the official repository for the OWASP Top 10 for Large Language Model Applications, now nested under the broader OWASP GenAI Security Project. It documents the most critical security risks facing LLM-based applications—prompt injection, data leakage, supply chain issues, and friends—tailored specifically to how these vulnerabilities manifest when there’s a chatbot in the loop.

The interesting bit

The project deliberately avoids copy-pasting the classic OWASP Top 10. Instead, it examines how conventional vulnerabilities mutate inside LLM architectures—where, say, “input validation” collides with natural language ambiguity and model behavior becomes a moving target. The v2.0 revision is currently in flight.

Key highlights

• Flagship OWASP project with a working group active on Slack (#project-top10-llm)
• Licensed CC BY-SA 4.0, so remix-friendly for training materials and internal docs
• Targets developers, data scientists, and security engineers building LLM plugins or applications
• Explicitly scoped to LLM application security, not the broader GenAI landscape (other frameworks exist for that)
• Maintained as a living document with community contributions via issues and PRs

Caveats

• The repository itself is a documentation hub; the actual guidance lives at genai.owasp.org
• “Currently scoped towards only LLM Application Security”—so don’t expect deep coverage of multimodal models or pure infrastructure hardening

Verdict

Worth bookmarking if you’re shipping anything with an LLM API behind it. Skip it if you’re looking for automated scanning tools or hands-on exploit code; this is awareness documentation, not a toolkit.