Building AI Agents That Attack Your AI and Build a Test Suite

Join our newsletter to stay informed about Innovate QA’s virtual and in-person events, opportunities, news, and offers.
Subscribe to the Innovate QA Newsletter

Speaker: Gregory Goldshteyn | LinkedIn

Topic: Building AI agents that attack your AI, and turning what they find into a test suite

Agenda:
🕛 12:00–12:05 PM PST — Welcome
🕛 12:05–12:55 PM PST — Gregory Goldshteyn | LinkedIn - Building AI agents that attack your AI, and turning what they find into a test suite
🕛 12:55–1:00 PM PST — Q&A & Closing

Summary:
AI applications can be tested with traditional methods, but adversarial testing becomes much harder when the system can reason, interact across multiple turns, and use external tools.

In this session, Gregory Goldshteyn will show how to build an AI attack agent that generates adversarial prompts, evaluates the target application's responses, learns from failed attempts, and continuously improves its attacks.

The session will explore how this approach can uncover issues such as prompt injection, sensitive information disclosure, and excessive agency. Gregory will also demonstrate how multi-turn attacks can gradually escalate toward a specific goal and why testing individual prompts may miss these behaviors.

The final step is turning those findings into something QA teams can use repeatedly. Gregory will show how successful attacks can become a regression corpus, how attack success rates can be measured, and how these tests can be replayed in CI whenever prompts, models, or tools change.

Takeways:
What attendees will learn:

1. How to build an AI attack agent using an attacker model, target application, judge model, and feedback loop
2. How to measure attack success rates across multiple trials
3. How to turn successful jailbreaks into repeatable regression tests
4. How multi-turn attacks can expose weaknesses that single-prompt testing may miss
5. How to test AI agents that interact with external tools
6. How indirect prompt injection can cause an AI agent to take unintended actions
7. How to incorporate AI security testing into practical QA workflows and CI pipelines

🎙️ Speaker Opportunities at Our Meetups

Have expertise in QA, leadership, or tech skills you'd love to share? We're always looking for engaging speakers for our monthly meetups. This is a fantastic platform to showcase your knowledge, connect with the community, and inspire others. If you're interested, submit your topic here — we'd love to hear from you!