IT Specialist Osman Demircan stated that AI assistants can perform tasks such as comparing products, tracking prices and making purchases based on instructions. Citing Amazon’s “Alexa for Shopping” assistant as an example, Demircan said that systems could be misdirected through “prompt injection” attacks and that user information could be transferred to unintended destinations. Demircan recommended that users check the product, seller, total cost, delivery address and subscription terms before payment.
Why it matters
The fact that AI assistants can do more than simply offer options during the shopping process and can carry out transactions based on instructions brings to the fore questions about which steps will be taken on behalf of users and how this authority will be limited. The possibility of prompt injection raises the risk that the assistant’s manipulation may not be limited to selecting the wrong product, but could also result in user information being transferred to unintended destinations. For this reason, the issue is important not only for consumers who use these systems, but also for services that manage payment and personal data security processes. Expert recommendations require the user to retain responsibility for the final decision and control in automated shopping; however, how assistants will technically support these checks and how erroneous instructions will be prevented remain open questions.