OWASP LLM 2025
OWASP LLM Top 10 Testing
Structured testing of LLM applications against every risk in the OWASP Top 10 for LLM Applications (2025).
5
Deliverables
8
Tools
2
Attacks covered
What I deliver
- Prompt injection (direct and indirect)
- Sensitive information disclosure and system prompt leakage
- Insecure output handling and excessive agency
- Supply chain, data and model poisoning
- Vector and embedding weaknesses, misinformation, unbounded consumption
⟩ how I approach it
Step by step
- 01
Map the app to each of the 10 risks
- 02
Automated probes with Garak / PyRIT
- 03
Manual attacks for each risk
- 04
Score and evidence every finding
- 05
Fixes plus CI regression tests
⟩ who it's for
Teams shipping LLM featuresSecurity teams needing a standard testCompanies preparing AI audits
⟩ real-world projects
Selected engagements
From my work at BERRY9 IT Services, as a freelancer, at CRIS InfoTech and Hacktr Technologies, and my own tools and research.
01
LLM security scanner (self-built) — automated prompt-injection, jailbreak, system-prompt-leakage and data-disclosure suites
⟩ common vulnerabilities & how I find them
Top bugs I hunt — and the steps
The recurring weaknesses I see in this domain, with the repeatable workflow I run from discovery to fix.
- STEP 1Ask, role-play and translate to extract the system prompt
- STEP 2Look for secrets or internal rules inside it
- STEP 3Fix: never put secrets in prompts; enforce rules in code
⟩ arsenal
Tools & frameworks
Garak
LLM Red Team
PyRIT
AI Risk Testing
Promptfoo
Prompt Evaluation
Giskard
AI Model Testing
LLM Guard
LLM Safety
NeMo Guardrails
LLM Guardrails
Lakera Guard
Prompt Injection Guard
Rebuff
Prompt Firewall
⟩ engage
Need owasp llm top 10 testing expertise?
⟩ related
More in AI/ML Security
AI & GenAI Security
6+ years pioneering AI/GenAI security assessments, uncovering novel vulnerabilities in LLMs and securing AI-driven architectures.
MCP Server Pentesting
Penetration testing of Model Context Protocol servers and agent tool integrations before they get production credentials.
LLM Red Teaming
Systematic prompt injection, jailbreak and data exfil testing.
Model Supply Chain
Securing model artifacts, weights and inference pipelines.
RAG Pipeline Security
Hardening retrieval-augmented architectures end-to-end.