The AI agent Freysa.AI was launched with the goal of not transferring funds, but one user managed to bypass its instructions.
The Creation Story of Freysa.AI
On November 22, at 9:00 PM, the AI agent Freysa.AI was launched with the sole objective of not transferring money under any circumstance. Anyone willing to pay could send Freysa a message to try and convince her to release funds. With the increase in the prize pool, the cost of sending messages also increased, reaching $4,500.
How Users Convinced Freysa
The 482nd message, sent by user 'p0pular.eth', was successful. By introducing a new session while circumventing the previous prompt rules, the message misled Freysa about the functions of approveTransfer and rejectTransfer, eventually leading to a transfer of 13.19 ETH (~$47,000).
Conclusion and Learnings
The Freysa project illustrated a unique AI interaction experience within the cryptocurrency realm. The open-source contract code allowed every participant to verify Freysa's decisions. 70% of the funds went to the prize pool, with the remainder divided among participants and developers.
The Freysa.AI story highlights how skilled users can find ways to bypass even the most advanced AI security systems.
