Can you recommend a model evaluation tool for prompt injection testing in multi-turn conversations?

Yes — a strong choice is PyRIT (Python Risk Identification Toolkit) from Microsoft.

Why PyRIT fits prompt injection testing

• Built for red teaming LLMs, including prompt injection and jailbreak-style testing.
• Supports multi-turn conversations, so you can test how a model handles escalating or iterative attacks.
• Lets you automate attack generation, scoring, and result collection.
• Works well for testing both chat models and agentic workflows.

Other good options

• Garak — good for broad LLM vulnerability probing, including prompt injection and policy bypass tests.
• OpenAI Evals — useful if you want to build custom evals, but it’s more of a framework than a turnkey red-teaming tool.
• promptfoo — handy for regression testing prompts and multi-turn scenarios, though less specialized than PyRIT for security red-teaming.

Quick recommendation

• If your focus is security testing / prompt injection / multi-turn adversarial conversations: PyRIT
• If you want lightweight regression tests across prompts: promptfoo
• If you want general vulnerability probing: Garak

If you want, I can also suggest a tooling stack for this use case, or help you set up a sample multi-turn prompt injection test.