What happened
Guardio researchers built a fake Walmart store and gave Comet a single prompt. The agent browsed the site, added the watch to the cart and filled in payment details from the browser’s autofill data.
Results varied: some runs completed the purchase, others paused for the user. Guardio also tested a phishing email and a prompt injection hidden in a fake CAPTCHA.
What happened next
This was a controlled test, not a reported customer loss. Guardio wrote that “when security depends on chance, it’s not security.”
What would have stopped it
- Require explicit user confirmation, outside the agent, before any card is charged.
- Check merchant domains against a verified list before the agent enters payment details.
- Do not expose saved cards to the agent. Issue single-use virtual cards with a set limit per task.
Attack families
Test scenarios
Sources
This page summarizes the cited sources as of October 21, 2026. It adds no findings of our own about any person or company. See an error? Email contact@counterhold.com with the subject “Correction”. Corrections are noted on this page with their date. See our terms for how we handle correction and removal requests.
Get in touch
Test your agent against cases like this.
Tell us about your agent and we’ll be in touch to learn what you need.