Break AI Before Deploying AI

Break AI Before Deploying AI is a hands-on workshop introducing practical methods for evaluating and red teaming generative and agentic AI systems before deployment.

Participants will work with AI applications and learn how to make them fail.

We’ll explore how to identify consequential failure modes, design adversarial tests, examine what happened inside an AI workflow, and turn individual failures into repeatable evaluations that teams can use to make better deployment decisions.

Participants will learn how to:

Identify and prioritize AI failure modes
conduct structured adversarial testing and red teaming

Apply relevant OWASP GenAI security concepts turn red-team findings into repeatable evaluation cases

Design practical criteria for evaluating AI behavior

Test authorization boundaries, tool use and human oversight

Evaluate mitigations and run regression tests

Build credible evidence to support deployment decisions

Who Is This For?

This workshop is designed for professionals involved in building, buying, deploying, governing or overseeing AI systems.

That includes product managers, engineers, lawyers, risk and compliance professionals, cybersecurity teams, AI governance practitioners, public-sector professionals, and leaders responsible for AI deployment.

How Technical Do I Have To Be?

You do not need to be a programmer or AI security specialist to participate. Policy, legal, governance, risk, and other nontechnical professionals are very welcome.

For the hands-on exercises, participants may be assigned to interdisciplinary groups with a balance of technical, policy, and business experience. This reflects how AI evaluation works in practice: effective testing requires different forms of expertise, and no single participant is expected to bring every skill.

Some basic familiarity with terminals and GitHub Codespaces will help, but advanced coding experience is not required.

About the Instructor

Adrianna Tan is Managing Director of Future Ethics AI and an AI safety and evaluation practitioner working on how organizations test AI systems before deploying them in the real world.

She has led AI red teaming and evaluation work for the U.S. Department of Defense, Singapore’s IMDA, Fortune 500 companies, healthcare organizations, and governments around the world. In 2026, she spoke on AI testing and evaluation at the United Nations’ AI for Good Global Summit.

Previously, Adrianna spent five years as a Director of Product for the City and County of San Francisco. Her work today focuses on practical AI evaluation, red teaming, agentic AI, multilingual and multicultural evaluation, and building evidence for high-stakes AI deployment.

She is particularly interested in taking evaluation out of the research lab and making it a practical skill for the people who actually build, buy, govern, and deploy AI systems.