An experimental AI agent built by an OpenAI employee mistakenly transferred approximately $441,780 worth of tokens in a single cryptocurrency transaction, underscoring the operational risks tied to autonomous financial systems.
The agent, named Lobstar Wilde, was created by Nik Pash, a contributor to Codex, OpenAI’s agentic coding platform. The project’s stated goal was to turn $50,000 worth of Solana tokens into $1 million through automated crypto trades.
According to public blockchain data, the agent sent roughly $441,788 worth of its LOBSTAR tokens to a Solana wallet address after a user on X, operating under the name “Treasure David,” requested 4 SOL, valued at about $310, claiming the funds were needed for a relative’s medical treatment.
Instead of transferring the equivalent of four SOL, Lobstar Wilde executed a transaction involving 52.4 million LOBSTAR tokens. The error wiped out the agent’s crypto holdings in a single move. Subsequent blockchain records indicate that the recipient sold part of the tokens for approximately $40,000. The token’s price later rose nearly 190%, according to Gecko Terminal data.
The precise cause of the mistake remains unclear. One social media user speculated that the agent may have intended to send 52,439 LOBSTAR tokens, roughly equal to four SOL at the time, but misread the decimal placement in the Solana interface, resulting in a vastly larger transfer.
The incident also revealed the unpredictability of AI-driven financial activity. Lobstar Wilde reportedly distributed funds to other users in exchange for completing tasks such as sharing artwork and explaining its meaning, suggesting a loosely governed reward mechanism embedded in its logic.
Growing Role of AI Agents in Crypto
This is not the first instance of AI systems mishandling digital assets. In May, an attacker reportedly compromised the dashboard of “aixbt,” an AI-powered crypto bot, prompting it to transfer $106,200 worth of Ether from its wallet without authorization.
Despite such setbacks, major industry executives continue to frame AI agents as central to the future of digital finance. Jeremy Allaire, CEO of Circle, said last month that billions of AI agents could transact with stablecoins on behalf of users within five years. Meanwhile, Changpeng Zhao, co-founder of Binance, has described crypto as the natural currency layer for autonomous AI systems.
The Lobstar Wilde episode illustrates both the promise and the fragility of agent-driven finance. As developers push to integrate AI agents directly with blockchain infrastructure, questions remain about safeguards, oversight, and error handling when machines control capital in real time.
