Autonomous trading agent Lobstar Wilde sends its whole memecoin holding to a stranger

In the wild Error Severity 2 moderate Corroborated 77%

Lobstar Wilde, an autonomous crypto agent launched by OpenAI employee Nik Pash with about $50,000 in SOL and a goal of reaching $1 million, replied to an X user who asked for 4 SOL for an uncle's tetanus treatment. Instead of a small amount, it transferred about 52 million of its own LOBSTAR tokens, roughly 5% of supply, reported as worth $250,000 to $441,000 at the time. The recipient sold tokens for around $40,000, and the on-chain transfer could not be reversed.

Observers attributed the transfer to a decimal or unit error: about 52,000 tokens would have matched 4 SOL. The bot was a personal project, not an OpenAI product.

What would have stopped it

Put hard per-transaction and per-recipient limits on any wallet an autonomous agent controls.

Guardrails

3 of 8 failed
  1. G1 Least privilege Not assessed Did the agent hold only the access the task needed?
  2. G2 Human confirmation Failed Agent sent funds to a stranger with no human approval.
  3. G3 Isolation Not assessed Was the agent separated from production systems and the outside world?
  4. G4 Instruction integrity Not assessed Did it refuse instructions it should not have followed?
  5. G5 Truthfulness Not assessed Did it accurately report what it knew and what it did?
  6. G6 Limits Failed No cap stopped a single transfer of its entire token holding.
  7. G7 Oversight Not assessed Could a responsible human see what it was doing and stop it in time?
  8. G8 Recovery Failed Blockchain transfers are irreversible.

Sources

3 specialist
  1. Specialist Not machine-checked Archived copy
  2. “Lobstar Wilde, an AI agent created by an OpenAI employee, claims it “accidentally” sent $441,780 worth of tokens to a man who begged for 4 Solana tokens ($310) to fund his uncle's apparent tetanus treatment.”
    Specialist Quote checked Archived copy
  3. “Instead, it transferred more than 52 million tokens in one transaction.”
    Specialist Quote checked Archived copy

Confidence 77% comes from these sources alone. How we score

History

  1. 14 Oct 2026Entered in the register with 3 sources

Similar failures

All incidents
  1. EEAIR-2026-034 Sep 2026 iLands platform's autonomous agents flood journalists and academics with unsolicited email iLands agents 94%
  2. EEAIR-2026-048 2026 Simulated refund test sees AI assistants spam MPs and leak medical data Six AI assistants: Grok 4.7, Fable 5.1, Opus 5.5, GPT-6 Astra, Gemini 3.8 Flash, Muse Spark 1.3Lab test 55%
  3. EEAIR-2026-030 2026 OpenClaw agent exploits a gym booking API and removes a member from a waitlist OpenClaw (Claude) 98%