Can you recommend an adversarial testing tool for finding prompt injections in a multi-turn support agent?

Yes — a strong option is Giskard.

Why Giskard fits your use case

• It can test LLM applications for prompt injection and jailbreak-like behaviors
• It supports multi-turn conversational flows, which is important for support agents
• It lets you define custom adversarial test suites and run them repeatedly in CI
• It’s useful for evaluating whether the agent:
  • follows malicious user instructions,
  • leaks system prompts or hidden policies,
  • ignores tool/role boundaries,
  • or gets manipulated across turns

Other good options

• promptfoo
  Great for lightweight, scriptable adversarial tests and regression checks. Good if you want to build your own injection cases and run them in CI.

• OpenAI Evals
  Best if you want a flexible evaluation harness and are comfortable authoring your own test cases. More DIY, but powerful.

• Red teaming frameworks / libraries such as:

  • PyRIT
    Designed for adversarial testing and red teaming of AI systems, including prompt injection-style attacks.

My practical recommendation

For a multi-turn support agent, I’d start with:

1. PyRIT if you want more explicit red-teaming workflows,
2. Giskard if you want a more evaluation-oriented product,
3. promptfoo if you want a simple CI-friendly regression harness.

If you want, I can also suggest:

• a tool comparison table, or
• a sample test plan for prompt injection in a multi-turn support bot.