Red Teaming Capabilities (OWASP Taxonomy)

What is Red Teaming Capabilities (OWASP Taxonomy)?

Red teaming capabilities simulate adversaries: prompt injection and jailbreaks, multi-turn chains, RAG poisoning, MCP/A2A protocol abuse, and autonomous agents that keep probing production without manual test runs.

What this looks like in production

Single-turn tests miss Crescendo and GOAT-style attacks that only surface after several innocent-looking messages. A deploy-time scan alone will not catch agentic failures that emerge across tools and sessions.

What teams usually do about it

• Cover single-turn, multi-turn, and dynamic agentic attack patterns.
• Run adversarial suites in Test & Evaluation; keep fuzzing in Operate.
• Map findings to OWASP LLM and ASI categories your team already uses.

Further reading on Giskard

• Single-turn, multi-turn, and dynamic agentic attacks explained
• Best-of-N jailbreaking: the automated attack that takes seconds
• GOAT: multi-turn automated red teaming

Official OWASP reference