Pen Test Partners' Artificial Intelligence Testing Service assesses the security of AI implementations across four deployment models: third-party AI integrations, cloud-based AI hosting, on-premises AI deployments, and custom AI deployments. The service targets risks specific to each deployment type, including supply-chain risk and data security for third-party integrations, and infrastructure security for on-premises deployments. Internal methodologies are aligned with industry standards such as OWASP and the OWASP Top 10 for LLM Applications 2025. Core testing areas include: **Prompt Injection Testing:** Evaluates AI model resilience against adversarial input manipulation, including attempts to override system prompts, modify metadata, or circumvent security policies through direct or indirect injection. **Training Data Security Assessment:** Reviews training datasets for sensitive information such as credentials, encryption keys, and PII, and assesses the effectiveness of data sanitisation techniques to prevent leakage via AI-generated responses. **Custom Scenario-Based Testing:** Supports bespoke testing goals, such as assessing whether internal users can access unauthorised data via AI, or whether chat logs are exposed to unintended parties on the network. When AI is embedded within a broader application or infrastructure, the service integrates with other testing disciplines (e.g., web application penetration testing), extending the methodology to address AI-specific risks in context rather than treating the AI as an isolated system.