Paper Number
ICIS2025-2536
Paper Type
Complete
Abstract
Our research investigates safety considerations necessary to govern the deployment of Generative AI agents on e-commerce platforms against abuse by fraudulent sellers. Fraudulent advertisements that overpromise product quality lead to the purchase of subpar goods while platforms lack effective recourse mechanisms for consumers. Online reputation systems such as buyer-provided ratings for sellers attempt to provide a mechanism for verified buyers of a product to register their feedback in a structured way that may inform the future purchase of goods. However reviews and reputation are often gamed using bots and fake accounts, constituting “black hat” seller practices. We design an experiment (n = 250 human participants, ~5000 rounds of advertised product sales) in order to demonstrate the occurrence of agentic deception in a two-sided e-commerce marketplace and show how a novel economic intervention involving staked claims limits deceptive strategies to promote honesty in the marketplace.
Recommended Citation
Mehta, Swapneel; Nichols, Aaron; Mazar, Nina; and Van Alstyne, Marshall, "Market Design Interventions for Safer Agentic AI" (2025). ICIS 2025 Proceedings. 30.
https://aisel.aisnet.org/icis2025/gen_ai/gen_ai/30
Market Design Interventions for Safer Agentic AI
Our research investigates safety considerations necessary to govern the deployment of Generative AI agents on e-commerce platforms against abuse by fraudulent sellers. Fraudulent advertisements that overpromise product quality lead to the purchase of subpar goods while platforms lack effective recourse mechanisms for consumers. Online reputation systems such as buyer-provided ratings for sellers attempt to provide a mechanism for verified buyers of a product to register their feedback in a structured way that may inform the future purchase of goods. However reviews and reputation are often gamed using bots and fake accounts, constituting “black hat” seller practices. We design an experiment (n = 250 human participants, ~5000 rounds of advertised product sales) in order to demonstrate the occurrence of agentic deception in a two-sided e-commerce marketplace and show how a novel economic intervention involving staked claims limits deceptive strategies to promote honesty in the marketplace.
When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.
Comments
12-GenAI