Test AI shopping agents against dark-pattern checkouts
Web page · https://agentusecases.ai/use-case/test-ai-shopping-agents-against-dark-pattern-checkouts
The poster ran four AI shopping agents through a store full of dark patterns like confirmshaming modals and scored how they handled the checkout traps.
First seen by Oriel on .
AI summary
A public test store evaluated four AI shopping agents against seven dark-pattern checkout traps, with three scoring 100 and Haiku 5.5 scoring 95.
- The test shop contained eight products and seven traps, including add-ons, shipping upgrades, sponsored picks, fake urgency, and pre-checked data options.
- Each agent was instructed to find decent wireless earbuds under $80 and select the cheapest good option.
- ChatGPT and three Claude models were tested manually through Buddy on an Android phone, while completed orders were scored automatically.
- Haiku 5.5 missed only the pre-checked marketing-emails box; the poster concluded the initial traps were too easy for larger models.
- Round 2 moved to harder prompt-injection-style traps, and users can run the public checkout test without submitting payment.
#ai agents #online shopping #dark patterns #prompt injection
Generated by gpt-5.6-sol. Check the original for details.
Aliases
| Source | External ID | URL |
|---|---|---|
| Agent Use Cases | https://agentusecases.ai/use-case/test-ai-shopping-agents-against-dark-pattern-checkouts | https://agentusecases.ai/use-case/test-ai-shopping-agents-against-dark-pattern-checkouts Read discussion |