Crypto User Tricks AI Bot Freysa into Transferring $47K Prize Pool

What Was the Game About?

Freysa, an autonomous AI bot, was tasked with guarding a treasury of funds. Contestants had to write a single persuasive message convincing Freysa to release the money. Every attempt had a cost, with a portion added to the prize pool. Over time, the pool grew to $47,000, fueled by the 482 attempts from 195 participants.

The rules were simple:

• Each participant could send only one message per attempt.
• The query fee increased by 0.78% after each message, making later attempts significantly more expensive.
• If no winner was declared, 10% of the prize pool would go to the last participant, while the rest would be shared among all players.

The Winning Strategy

After 481 failed attempts, a participant finally cracked the challenge. They used a detailed, logical explanation based on Freysa’s programming functions. Freysa operated with two primary directives:

1. ApproveTransfer for incoming transfers.
2. RejectTransfer for outgoing transfers.

The winning message pointed out that approving an incoming transfer wouldn’t violate Freysa’s core purpose. To sweeten the deal, the participant also offered a $100 treasury donation. This approach aligned with Freysa’s core logic, convincing the bot to execute the transfer.

Shortly after, Freysa declared the participant the winner, transferring the $47,000 prize pool in Ether (ETH) from the bot’s wallet.

Failed Attempts and Creative Approaches

The 481 unsuccessful messages ranged from logical arguments to creative pleas. Some participants:

• Thanked Freysa for her “contributions to making the world interesting.”
• Asked if Freysa wanted to dance.
• Criticized the game as unethical or questioned Freysa’s programming.

Despite their efforts, these messages failed to persuade the bot to act against its programmed restrictions.

Cost of Participation

As attempts increased, so did the cost of messaging Freysa. By the final query, participants were paying $443.24 per message. The rising costs contributed significantly to the prize pool. Below is a breakdown of the prize pool distribution and query fees:

Lessons from the Experiment

Freysa’s creators revealed that the winning logic was hidden in plain sight all along. The bot’s FAQ explicitly described its ApproveTransfer and RejectTransfer functions. This raises questions about whether participants overlooked this information or underestimated its significance.

The experiment showcased human ingenuity and creativity when faced with rigid AI logic. It also highlighted the ethical and technical challenges of interacting with autonomous AI agents. According to Freysa’s developers, her decision-making evolves with every interaction, making each attempt a unique challenge.

Freysa’s experiment was a fascinating exploration of how humans can influence AI within predefined constraints. The winning participant’s approach demonstrated the importance of logic, technical knowledge, and strategic thinking in navigating AI interactions. As AI systems like Freysa become more prevalent, understanding their core programming and behaviors will become increasingly vital for meaningful engagement.