CBRX AI Red Teaming is a service that performs offensive security testing on AI systems including LLM applications, agents, RAG systems, and AI supply chains. The service simulates attacker behavior to identify vulnerabilities and failure modes in AI deployments. The service tests for multiple attack vectors including prompt injection and jailbreaking attempts that override instructions, leak secrets, or bypass policies. It evaluates data exfiltration risks and privacy failures where sensitive or proprietary data can be extracted through model queries. The testing covers agentic systems to identify manipulation vectors that could lead to harmful actions, and RAG systems to detect retrieval poisoning and document manipulation vulnerabilities. The service also assesses AI supply chain weaknesses across third-party models, plugins, gateways, APIs, and integrations. Deliverables include a red team report documenting attack paths with impact and likelihood assessments, reproducible attack scenarios, prioritized remediation recommendations covering prompts, access control, logging, and guardrails, and secure architecture recommendations. The service targets organizations with live LLM applications or pre-production pilots, CISOs requiring AI security due diligence, and AI or product teams deploying rapidly.