Fraudulent sellers are a growing problem, especially for the secondary market, prompting research on the safe use of (Generative) AI agents in online marketplaces. These AI systems have the potential to improve efficiency and personalize services, but they also introduce new risks if misused or poorly managed. Online shoppers sometimes buy low-quality used and refurbished products because sellers exaggerate product claims or use misleading advertisements. Platforms often cannot fully protect buyers, leaving consumers at risk. Ratings and reviews are intended to provide guidance, but fake accounts and automated bots frequently undermine these systems, making it difficult for genuine buyers to know which sellers are trustworthy. This issue highlights the urgent need for more reliable safeguards and oversight.
What is Agentic AI
Generative AI combined with autonomous, goal-driven behaviour is creating Agentic AI systems. These agents can handle complex tasks on their own, from researching products to completing transactions. While they save time and reduce human workload, they also raise safety and trust concerns. Mistakes or misuse in high-stakes areas like finance, healthcare, and e-commerce could have significant consequences, emphasizing the need for clear rules and monitoring.
How market design can help
New research (Market Design Interventions of Safer Agentic AI) suggests practical methods to make these AI agents safer. Interoperability protocols, such as the Agent-to-Agent Protocol (A2A), allow agents from different companies and architectures to work together securely without constant human supervision. This reduces errors, ensures smooth coordination, and lays the foundation for marketplaces where multiple autonomous agents can interact safely.
Making transactions accountable
The Agent Payments Protocol (AP2) links a user’s intent, shopping cart, and payment with a secure digital record. This creates a verifiable, traceable, and transparent system for automated transactions, ensuring that purchases are accountable and reducing the risk of disputes or fraud. By integrating these protocols, e-commerce platforms can maintain trust while allowing agents to operate independently.
Governance and oversight
Organizations integrating these systems must define clear rules for how much freedom agents have, which decisions they can make, and how to monitor their actions. Human oversight remains essential, particularly during early deployment. Checking agent decisions builds reliability, prevents misuse, and fosters trust. Over time, this approach allows for safe scaling of fully autonomous operations.
Understanding Generative AI agents
A Generative AI Agent (GenAI Agent) combines content creation, such as text, images, or software, with autonomous decision-making. These agents do more than respond to prompts. They plan tasks, remember past actions, reason about next steps, and interact with external tools or applications to achieve specific goals efficiently and independently.
Key functions of AI agents
Planning breaks a goal into clear, actionable steps. Memory enables the agent to learn from past interactions, improving consistency. Tool use connects the agent to applications, APIs, or databases, allowing it to retrieve information or act in real time. Reasoning helps the agent decide the best course of action and produce desired outcomes. Combined, these capabilities allow agents to manage complex tasks like marketing campaigns, financial reports, or product research autonomously.
Safer AI for e-commerce
By implementing clear rules, monitoring systems, and secure protocols, e-commerce platforms can safely benefit from Generative AI agents. These agents can improve efficiency, offer personalized experiences, and enhance operational scalability while maintaining consumer trust. With proper governance, platforms can ensure that autonomous AI acts responsibly, traceably, and aligned with business objectives.
Market
Trade-in
Repair
Refurbishing
