AI Security (LLM Top 10 v2025)

What Gets Tested

OWASP LLM Top 10 v2025 vulnerability testing

Prompt injection detection (direct & indirect)

System Prompt Leakage probes (LLM07, NEW in 2025)

Vector & Embedding RAG poisoning checks (LLM08, NEW in 2025)

Unbounded Consumption / denial-of-wallet testing (LLM10)

Insecure output handling checks

Multi-turn / multimodal red-team via Microsoft PyRIT

Model Denial of Service testing

Supply chain vulnerability assessment (AI dependencies)

Sensitive information disclosure via AI responses

Excessive agency / permission testing (LLM06)

Model extraction attempt detection

AI endpoint rate limiting assessment

MCP server discovery + CVE-2025-49596 risk-class probe

Findings mapped to MITRE ATLAS technique IDs

1

Identify AI/ML endpoints, chat interfaces, MCP servers, API integrations.

2

Systematic prompt injection via garak + promptfoo + PyRIT multi-turn / multimodal converters.

3

Check for sensitive data leakage, PII exposure, system-prompt leakage, hallucination risks.

4

Rate limiting, resource limits, denial-of-wallet resilience, MCP transport sandboxing.

OWASP LLM Top 10 v2025 MITRE ATLAS v5.4 EU AI Act OWASP-A03 (Injection)

€89

Any tier · One-time per scan

151 specialized tests + AI-powered analysis

Select this addon when configuring your scan

85% of real breaches exploit authenticated access — test what logged-in users can do

Full exploitation testing — SQL injection, RCE, credential brute force, consent-gated proof-of-exploit