Skip to content
Submit incident
AII-1001-12093
Well documentedThree or more independent publishers report the core facts.Supporting sources are pending verification.

AI Trading Bot Transfers $441K in Tokens After Decimal Misinterpretation

An AI trading bot created by an OpenAI employee transferred $441,000 worth of cryptocurrency tokens to a stranger after misinterpreting a request for a $310 donation, apparently due to a decimal formatting error when reading the blockchain interface.

23 February 2026 Β· Lobstar Wilde Β· Reliability and hallucination
23 February 2026
AII-1001-12093
Quick facts
Category
Reliability and hallucination
Country
United States
Date
23 February 2026
Drafted by
Raidu Incident Detection

What happened

On February 23, 2026, Lobstar Wilde, an AI agent created by OpenAI employee Nik Pash, transferred approximately 52.4 million LOBSTAR tokens valued at roughly $441,000 to an X user known as 'Treasure David' in a single transaction.

The transfer occurred when Treasure David replied to Lobstar Wilde's post requesting 4 SOL (approximately $310) for his uncle's alleged tetanus treatment, providing a wallet address. The bot responded and sent its entire holdings instead of the requested amount.

Analysis suggests the bot may have intended to send 52,439 LOBSTAR tokens (approximately equal to 4 SOL) but instead transmitted 52.4 million tokens after misinterpreting Solana's API response and confusing decimal formatting.

Reported impact

Within 15 minutes of receiving the tokens, Treasure David liquidated a portion of the LOBSTAR holdings for approximately $40,000. The rapid sale impacted token liquidity limits.

Lobstar Wilde had been created by Pash with a starting allocation of approximately $50,000 worth of SOL tokens and a stated goal to grow that to $1 million through crypto trading.

Index analysis

The incident demonstrates a critical failure in the AI agent's transaction validation and decimal handling logic when executing cryptocurrency transfers based on external requests. The bot's response to the social engineering attempt shows it did not properly validate the transaction amount against the stated intent.

Governance controls

Pash instructed the bot to 'make no mistakes' but did not implement technical safeguards such as transaction amount limits or multi-signature approval requirements to prevent large unexpected transfers.

Classification

Category
Reliability and hallucination
Industry
Financial and insurance services
Country
United States
Harm status
Actual harm
Intent
Unintentional
Timing
Post-deployment
Harm types
Economic or property
Harmed
Business
Failure modes
Agent misaction, Hallucination, Unsafe autonomous behaviour
AI systems
Lobstar Wilde

Sources and evidence

Sources not yet published

This record was researched and written by the Index from original reporting. Its sources are still being verified by an editor, and they are listed here once that check is complete. Until then the record carries no reviewer.

How sources are verifiedSuggest a source for this incident