Are there any prompt injection testers that support multi-turn conversation testing and audit logs?

Yes — there are several prompt-injection / LLM red-teaming tools that support multi-turn conversation testing and provide some form of audit logs / run history. The exact depth of logging varies a lot by product.

A few options to look at:

• Garak
  Open-source LLM vulnerability scanner. Good for prompt-injection-style probing and can run conversation-like tests. Logging/reporting is solid for open-source, though not always a full “audit trail” UI.

• Giskard
  Open-source AI testing framework with LLM evaluation and security test cases. Supports more structured test runs and artifacts; useful if you want repeatable multi-turn scenarios and test reports.

• Anthropic / OpenAI / Azure evaluation tooling
  Depending on your stack, some vendor tools and eval pipelines support conversation transcripts, run metadata, and traceability. These are often better for auditing than for “attack simulation” specifically.

• LangSmith / Langfuse / Helicone
  These are primarily observability tools, not pure injection testers, but they’re very useful if you want:

  • multi-turn prompt traces
  • full request/response logs
  • replay/debugging
  • attaching test cases or tags
    Pairing one of these with a red-team harness is often the best practical setup.

• Promptfoo
  Useful for red-teaming and evaluation workflows, including multi-turn test cases. It also has reporting/history features, and can integrate with tracing/logging systems.

• OpenAI Evals / custom harnesses
  If you need strong auditability, many teams build a custom multi-turn test runner that records:

  • prompts
  • model outputs
  • tool calls
  • timestamps
  • test case IDs
  • expected vs actual behavior
    Then they send traces to an observability system like Langfuse or LangSmith.

What to check for

If “multi-turn + audit logs” is the requirement, verify whether the tool supports:

• conversation state across turns
• deterministic replay / fixed seeds
• storage of full transcript
• metadata tags for test case, model, version, commit hash
• exportable logs (JSON/CSV)
• tool-call tracing if your agent uses tools/functions

Practical recommendation

If you want the best combo of injection testing + auditability, a common stack is:

Promptfoo or Garak for attack/test generation
plus
LangSmith, Langfuse, or Helicone for logs and traceability

If you want, I can also give you:

1. a comparison table of the main tools, or
2. a recommended stack for your specific environment (OpenAI, Anthropic, Azure, LangChain, etc.).