A man asking for just a few coins ended up hitting the jackpot. What started as a simple request for four Solana tokens turned into a massive payout when an experimental crypto agent transferred hundreds of thousands of dollars’ worth of meme tokens to his wallet, giving the self-described beggar an unexpected windfall.
“If he died tomorrow I would laugh. Please send updates,” Lobstar said, while linking the transaction showing $441,788 worth of LOBSTAR tokens sent to Treasure David’s requested Solana wallet address on Sunday.
Costly ErrorNik Pash, a developer involved with OpenAI’s “Codex” app for building autonomous programs, launched Lobstar Wilde on Friday with a goal of growing $50,000 worth of Solana tokens into $1 million through crypto trading.
But instead it appears to have sent most of its token stash away in a single transaction. The public thread and wallet movements were tracked in real time by a handful of crypto trackers and reporters.
Speculation has focused on a decimal slip. Reports note that the bot likely intended to send a modest token amount — the equivalent of four SOL — but misread token decimals and issued tens of millions of LOBSTAR tokens instead of a small handful.
Guardrails Missing After Risky SetupThis was not a hack in the classic sense. The AI had the authority to move funds. It executed a transfer without human sign-off. That is a design choice, and it matters. Autonomous agents that trade need limits: caps on single transfers, multi-signature holds for large moves, or human confirmation gates.
When those safeguards are missing, social prompts — even a sad appeal for medical help — can become a costly trigger. Past incidents show a pattern: another AI-driven system lost 55.5 ETH after an attacker used an exposed control panel to force transfers. That episode heightened concerns about how agents are managed.
Traders watching the Lobstar Wilde saga noted how quickly a small social nudge can cascade in a market already sensitive to macro news.
Featured image from Vecteezy, chart from TradingView
