Using AI: Amazon app searches and orders in third-party online stores
Amazon also wants to enable its users to shop on third-party websites. Customers do not have to leave the Amazon app to do this; an AI orders for them.
(Image: Nicole Glass Photography/Shutterstock.com)
If a product they are looking for is not available on Amazon, customers should not have to buy elsewhere in future: An AI will search for them in third-party online stores and, if desired, order there on their behalf. "Buy for me" is the name of the new AI function, which is now available as a beta version for some US Amazon customers.
If you search for products in the Amazon Shopping app, you will also be shown offers that other online stores have in their range. They can be called up and viewed in the Amazon app in the same way as products offered directly by Amazon. An "AI shopping agent" orders from the third-party provider on behalf of the Amazon user. All customers have to do is place their order in the Amazon app as usual. In a blog post, Amazon shows the process of such an order on the smartphone.
Amazon AI operates web browser
Behind the scenes, the shopping agent visits the external website, selects the desired product, enters the user's name, delivery address and payment details in encrypted form and completes the purchase, Amazon explains. The agent runs with Amazon's own Nova AIs and the Claude AIs from Anthropic. Nova Act, a spin-off that Amazon unveiled this week, is likely to be behind the new function. It can control web browsers and is already capable of simple work steps on external websites, reports tech magazine Techcrunch. Nova Act is already available to developers on Amazon.
However, "Buy for me" is also likely to raise some concerns. After all, it is still an AI that places orders on behalf of Amazon users and processes sensitive data such as bank details, home addresses, etc. And AIs are known to have a tendency to hallucinate and are prone to errors. The former, for example, leads to seemingly correct results from the AI, which may be objectively incorrect and cannot be plausibly traced back to the training data used. In the case of Amazon's shopping agents, for example, the AI could suddenly turn ten ordered items of a product into 1000. Techcrunch also reports that AI shopping agents still often take a very long time to process orders and frequently fail to do so. OpenAI and Google already offer similar 'shopping agents', but users still have to enter their data themselves.
Videos by heise
Some customers are also likely to be concerned about the security of their data. Jailbreak methods, for example, are becoming known time and again. This allows the security guidelines of AIs to be overridden so that they suddenly reveal sensitive data. For example, heise online was recently able to get the Claude AI to talk about building instructions for Molotov cocktails and other things. Only some of the answers were hallucinated. Later, Claude also revealed a list of identities of its developers, which were also not hallucinated. In a normal state, Claude refuses such requests. Chatbots based on large language models (LLMs) are susceptible to jailbreaks. Whether data from Amazon's AI shopping agents could also be read in this or a similar way is of course questionable. It remains to be seen whether this will cause reservations among Amazon users, and whether these are justified.
(nen)