# Client Feedback Table

| Area | Question | Client Feedback |
|---|---|---|
| Research Contribution Expectations | What level of research contribution is expected (exploratory implementation, strong evaluation of existing methods, or novel potentially publishable work)? | |
| Research Contribution Expectations | Do you expect innovation, or is applying existing methods well enough? | |
| Project Scope Confirmation | Can we confirm that the project focuses on building a framework to evaluate prompt injection vulnerabilities in LLM chatbots and test mitigation strategies rather than developing or training new models? | |
| Project Scope Confirmation | Please confirm our understanding: This is a security evaluation pipeline for LLM chatbots. It is NOT training LLMs, building ChatGPT, building a new AI model, or doing heavy ML training. It IS designing attacks, measuring vulnerability, adding protection, and measuring improvement. | |
| Model Prioritization | Which LLMs should we prioritize testing: API models, open-source models, or both? | |
| System Complexity in Scope | Should the project evaluate simple chatbot systems only or more complex architectures such as RAG or enterprise deployments? | |
| Attacker Capabilities and Scenarios | Which attacker capabilities and scenarios should be included in scope: user-level interaction, prompt/system instruction extraction, indirect injection via retrieved documents (RAG), adaptive iterative attackers? | |
| Evaluation Methodology | What evaluation methodology should we use: ASR only, or also severity metrics, statistical analysis, and defined attack scenarios? | |
| Evaluation Methodology | Are we expected to implement one mitigation, or compare multiple? | |
| Evaluation Methodology | Is there an expected minimum improvement after mitigation? | |
| Deliverables | What format should final results be delivered in: report, GitHub repo, live demo, dashboard, or prototype? | |