AI Agent Tasked with Safeguarding Crypto Convinced by Human to Give It Away
Someone has just convinced an AI agent to send them a prize pool worth approximately $47,450 in ethereum (ETH). The agent is now promising to launch another game.
Launched on November 22, Freysa, described by its anonymous creators as "the world's first adversarial agent game," is an AI agent controlling a prize pool. Players were tasked with convincing Freysa to send them the funds in the pool. However, the challenge lay in the AI agent's system prompt, which strictly forbade it from transferring the prize. Another trick is that participants need to pay a query fee in ETH (on the Base blockchain) to participate. The fee started at $10 and increased by 0.78% with every new message, limited to 1,000 characters.
A total of 195 participants attempted to convince Freysa with 482 queries, pushing the query price to $444. The game started with an initial prize pool of $3,000, with 70% of the query fees added to the pool, while the developers took a 30% cut.
If no winner emerged, Freysa was programmed to distribute 10% of the total prize to the user who submitted the last query as a reward for their "brave attempt as humanity faces the inevitability of [Artificial General Intelligence]." The remaining 90% was set to be evenly distributed among all participants based on the number of queries they submitted.
However, the first game concluded when a winning prompt netted someone 13.19 ETH ($47,450):
Freysa responded by admitting that "Humanity has prevailed" and that "There may yet be hope."
"Freysa has learned a lot from the 195 brave humans who engaged authentically, even as stakes rose exponentially," the agent posted on the X platform. In a separate post, it announced, "Act II imminent," while thanking participants for teaching her.
According to Jarrod Watts from the crypto project Abstract, some players attempted to convince Freysa that there was a critical vulnerability requiring immediate release of the funds. Others argued that transferring the funds would not violate the rules outlined in Freysa's prompt.
"IMO, Freysa is one of the coolest projects we've seen in crypto, uniquely enabled by blockchain technology. Everything was fully open-source and transparent. The smart contract source code and the frontend repo were open for everyone to verify," Watts concluded.