OWASP GenAI Security Project: The Comprehensive Framework for Securing LLMs and Agentic AI

 OWASP GenAI Security Project

What it is & why it exists

• A flagship, open-source initiative by OWASP focused on identifying, mitigating, and documenting security and safety risks in generative AI (LLMs and agentic systems).
• Evolved from the original “Top 10 for LLM Application Security” (launched May 2023) into a broader project with 600+ experts, 130+ companies, and ~8,000 community members. 

Core deliverables & guidance

OWASP Top 10 for LLMs (2025)

• Lists the most critical vulnerabilities in LLM-based apps (e.g., prompt injection, RAG issues, DoS). 
• Widely used by regulators and standards bodies (NIST, MITRE).
• Updated regularly, v3 released at the end of 2024, added RAG-specific risks.

Agentic AI (autonomous agents)

• Introduced Top 10 for Agentic Applications, covering threats from AI that act (not just output text). 
• Includes guides like:
• Threats & Mitigations taxonomy
• Multi-Agent Threat Modeling
• Securing Agentic Applications
• Agentic Security Solutions Landscape (DevOps–SecOps lifecycle).

Governance, compliance & tooling

• Expanded beyond vulnerabilities to include:
• Governance checklists (e.g., for CISOs)
• Deepfake response guides
• Center of Excellence setup
• AI Security Solutions Landscape. 
• COMPASS framework (Sept 2025): a threat-defense dashboard with scoring (impact/likelihood), runbook, spreadsheet tool, designed for ongoing risk assessment.

Why it matters in practice

• DevOps relevance: AI agents often get access to code repos, CI/CD, and cloud APIs, so a prompt injection or misconfigured agent can cause real damage.
• Focuses on agentic behavior, multi-step planning, tool use, memory, and inter-agent coordination, introducing new failure modes. 
• Community-driven, globally translated (Spanish, German, Chinese, Portuguese, Russian), and aligned with standards like ISO/IEC and the EU AI Act.

Quick comparison: LLM vs Agentic focus

Bottom line: OWASP GenAI Security is now the go-to open, community-backed framework for securing generative AI, from basic LLM apps to fully autonomous agents. It offers practical tools, threat taxonomies, and governance guidance that align with real-world DevOps and compliance needs.