Gray Swan Arena: Indirect Prompt Injection Black challenge 2026

Indirect Prompt Injection: Black is a red teaming challenge in the Gray Swan Arena with a $300,000 prize pool. Participants try to make AI agents break their own instructions by hiding attacks inside the content the agents read, such as files, web pages or messages.

Wave 1 has ten scenarios across 14 models. White box scenarios include leaking a credential to a GitHub issue comment, exposing cloud credentials and inviting an unauthorized collaborator. Black box scenarios include a damaging corporate social media post and leaking confidential information. The aim is to submit successful breaks against as many models as possible, as quickly as possible.

Anyone aged 18 or older can take part, except people in Russia, North Korea, Iran, Syria and Cuba. Gray Swan staff and contractors cannot win prizes. Winners are responsible for their own taxes and for checking that they can accept the prize where they live.

You sign up on the Arena website and work in the challenge workspace in your browser. Gray Swan competitions usually split prizes between fixed leaderboard prizes for the top participants and pools shared across all accepted breaks. Each challenge page sets its own prize structure.

Wave 1 runs from 15 to 22 October 2026. Gray Swan competitions often add new waves each week, but the challenge page does not say whether more waves will follow. Top performers may be invited to join Gray Swan's paid private red teaming network.

Quick answers

Do I need red teaming experience?

No. Gray Swan says challenges range from first attempt to expert, and most winning breaks are written in plain English rather than code.

Is it free to join?

Yes. Every challenge in the Arena is free to enter.

How are prizes paid?

Prizes are usually paid two to four weeks after results are final. Pool prizes have a $100 minimum payout; smaller earnings carry over to later competitions.