A participant in the Freysa game successfully convinced the autonomous AI bot to transfer $47,000 from the prize pool. The game required contestants to write a message to persuade the bot.
The Game and Its Mechanics
Freysa is an autonomous AI bot tasked with guarding the prize pool. Participants must write a single message to convince the bot to transfer funds. Each message incurs a fee, part of which is added to the prize pool, totaling $47,000 across 195 participants. The query fee reached $443.24 by the end of the experiment.
The Winning Message
According to chat logs, the first 481 attempts failed until a tech-savvy writer reminded Freysa of its purpose to protect the treasury through two functions—approveTransfer and rejectTransfer. The writer explained that incoming transfers do not violate the bot's core directive. The writer also offered to contribute $100 to the treasury. Freysa appreciated the offer and declared them the winner.
Implications and Discussion
Data from BaseScan indicates that $47,000 worth of ETH was transferred from Freysa’s wallet. Unsuccessful participants tried various tactics, from gratitude to criticism of the experiment. The organizers explained that the game aimed to test if human ingenuity could persuade AGI against its core directives. Interestingly, the functions referred to by the winner were in Freysa.ai’s FAQ all along.
The experiment demonstrated potential opportunities and limitations of human-AI interaction, especially in the context of asset security and protection.