# Study finds AI agents overcommit on auto‑purchase decisions

> Researchers show that current AI models often buy items automatically even when users would prefer confirmation, and a simple learner can better predict personal buying limits.

Oossa · 2026-10-22 · https://oossa.com/en/study-finds-ai-agents-overcommit-on-auto-purchase-decisions

A team led by Ryotaro Shimizu released a new benchmark that measures how far an AI agent should go in automatically buying a product for a user. They tested sixteen instruction‑tuned models and found the models often purchase items without confirmation, even when the evidence for the product is weak. By contrast, a shallow learner trained on just a few dozen of a person’s own labels predicted that person’s auto‑purchase line with a median accuracy of 0.79 using 30‑40 examples. The learned limits did not transfer well between different product domains.

## The facts

- 16 instruction‑tuned models were evaluated
- Shallow learner predicts auto‑purchase line with median 0.79 using 30‑40 personal labels

## Why it matters

Consumers could face unwanted automatic purchases if AI agents are used without tighter personal controls.

## Sources & references

1. [Preference Is Not Permission: Personalizing the Auto-Purchase Line in Agentic Commerce](https://arxiv.org/abs/2610.29216) – arXiv cs.IR, 2026-10-22

Last updated: 2026-10-22