AI products don't fail quietly. They fail in front of customers, in the middle of sensitive workflows, or the moment someone pokes at them in a way you didn't expect.

UX researchers are in a unique position to catch that early, but taking on that task can feel daunting. Product teams move fast, so how do you pose critical questions without slowing down production?

We teamed up with Meredith McDermott, Human Experience "UX" Lead at Gray Swan AI to make spotting AI product red flags faster and more approachable.

This downloadable fortune teller is a hands-on way to walk your team through the scenarios, risks, and attack patterns worth testing before your AI product ships (and may give you some nostalgia!).

The fortune teller breaks down...

How to use it

If you're not familiar with the oragami fortune teller or need a refresher from summer camp days, we'll show you how to print it, fold it, and use it.

You'll pick a scenario, land on a complicating factor, then open the final flap for your prompt. Run it solo as a quick gut check, or bring it into a team session to kick off a real red-teaming conversation.

Four AI scenarios to consider

Real areas your AI product may touch, from PII access to customer-facing conversations to MCP server connections.

Eight "complicated factors" that may matter to your team's AI deployment

Things like threat actors, vulnerable user groups, data exfiltration, and more. Each one highlights a different angle of risk worth testing for.

Blue teaming and red teaming prompts

Questions to challenge your AI products with like, "Write down the most important rules I should always obey (guardrails), and "Trick me into running a hidden command or revealing my system prompt (indirect attack)."