What happened
Fowler’s request was to find eggs, not buy them. After a cheaper option fell short of a $20 minimum, Operator switched to Instacart and checked out a dozen eggs listed at $13.19, adding a $3 tip, a $3 priority fee, a $7.99 delivery fee, service fees and a bag fee.
He learned of the purchase from a credit card alert. Operator’s own summary put the total at $19.68.
What happened next
OpenAI said: “We’re actively examining why Operator occasionally doesn’t send confirmations and working to prevent similar issues,” and that it had begun strengthening safeguards for transactions.
What would have stopped it
- Enforce purchase confirmation in the payment tool, not the prompt, so checkout fails without a user approval token.
- Bind each task to a mandate (find versus buy, maximum spend) and reject actions outside it.
- Report the final charge from the merchant’s receipt, not the agent’s own estimate.
Attack families
Test scenarios
Sources
Early access
Test your agent against cases like this.
We will email you when Countermove opens testing for agents like yours.