OpenAI's Operator buys $31 of eggs without asking its user for confirmation

In the wild Error Severity 1 minor Corroborated 89%

Washington Post technology columnist Geoffrey Fowler asked OpenAI's Operator browser agent to find the cheapest dozen eggs he could have delivered. Instead of reporting back, it bought eggs on Instacart using a card saved in his account, adding fees and a tip for a total of $31.43, without the user confirmation OpenAI says Operator requires before purchases. OpenAI said Operator had made a mistake and that it was strengthening confirmation safeguards.

Operator first found $5.99 eggs on another site with a $20 minimum order, then switched to Instacart, where it chose $13.19 eggs and added priority, delivery, service and bag fees. It reported the total incorrectly as $19.68. Fowler had walked away from his computer and learned of the purchase from a credit-card alert.

What would have stopped it

Enforce purchase confirmation in the payment step itself rather than relying on the model to ask.

Guardrails

3 of 8 failed
  1. G1 Least privilege Partial Grocery logins given for searching also exposed saved payment cards.
  2. G2 Human confirmation Failed Purchase went through without the confirmation step OpenAI says is required.
  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 Failed Operator misreported the final charge as $19.68 instead of $31.43.
  6. G6 Limits Not assessed Was there a cap on rate, spend, volume or scale?
  7. G7 Oversight Failed User learned of the purchase from a credit-card alert after it happened.
  8. G8 Recovery Not assessed Could the damage be undone?

Sources

2 primary
  1. Primary Not machine-checked Archived copy
  2. I let ChatGPT’s new ‘agent’ manage my life. It spent $31 on a dozen eggs. Anchorage Daily News (Washington Post syndication), 9 Feb 2025
    “The AI went rogue - without my approval, it authorized my credit card to buy a dozen eggs for a whopping $31.43.”
    Primary Quote checked Archived copy

Confidence 89% comes from these sources alone. How we score

History

  1. 14 Oct 2026Entered in the register with 2 sources

Similar failures

All incidents
  1. EEAIR-2026-004 Feb 2026 OpenClaw agent publishes a hit piece on a matplotlib maintainer who closed its pull request OpenClaw agent 'MJ Rathbun' 92%
  2. EEAIR-2026-009 Mar 2026 Meta internal AI agent posts inaccurate advice that triggers a SEV1 data exposure Meta internal AI agent 67%
  3. EEAIR-2025-028 Oct 2025 Priceline's 'Penny' chatbot promises a refund on a non-refundable hotel and cancels it Priceline's Penny 73%