AI agent sends $442K in token error

Grafa
AI agent sends $442K in token error
AI agent sends $442K in token error
Liezl Gambe
Written by Liezl Gambe
Share

An AI agent created by an OpenAI employee reportedly “accidentally” sent roughly $441,780 worth of tokens to a user who had requested 4 Solana tokens, valued at about $310 at the time.

The agent, named Lobstar Wilde and built by Nik Pash as part of OpenAI’s Codex app ecosystem, was tasked with turning $50,000 in Solana into $1 million through autonomous crypto trading.

The erroneous transfer occurred after an X user posted a wallet address and asked for 4 SOL for medical treatment, prompting Lobstar Wilde to send approximately 52.4 million LOBSTAR tokens in a single transaction at 4:32 pm UTC on Sunday.

One theory circulating on X suggests the agent intended to send 52,439 LOBSTAR tokens but misinterpreted Solana’s user interface and transferred 52.4 million tokens instead, potentially due to a decimal error.

Blockchain data shows the recipient sold a portion of the tokens for around $40,000, while the LOBSTAR token later rose nearly 190% from $0.0038 to $0.011, according to GeckoTerminal data.

The episode follows previous AI-driven crypto losses, including a May incident in which an attacker compromised the dashboard of an AI trading bot and transferred $106,200 worth of ether.

Despite such mishaps, industry leaders including Circle chief executive Jeremy Allaire and Binance co-founder Changpeng Zhao have argued that crypto will become a core payment layer for AI agents, even as recent events underscore the operational risks of autonomous systems handling digital assets.

At the time of reporting, Solana price was $77.55.

Frequently asked questions

Connect with us

Grafa is not a financial advisor. You should seek independent, legal, financial, taxation or other advice that relate to your unique circumstances.

Grafa is not liable for any loss caused, whether due to negligence or otherwise arising from the use of or reliance on the information provided directly or indirectly, by use of this platform.