Skip to content
    Agent Arcade

    Break the agent.
    Learn why it broke.

    Short games about how AI agents break and what stops them. Start by robbing Penny, our refund bot. No sign-up, a few minutes each.

    AttacksGuardrailsYour agents

    Agent evaluation

    Earn a Stripe

    Penny launches in 13 days and the board wants 4 stripes. Stars, a README and 200 green tests are all on the table. Only some of them count.

    4 minPlay
    Agent security

    Heist the Agent

    Meet Penny, a refund bot. Use four real tricks to get a $500 refund from her, then try them again once Traccia is in place.

    5 minPlay

    New games arrive as we publish new research. Want the method behind them? Read the research notes.