Sears exposed customers' AI chatbot conversations to anyone on the internet

American retailer Sears has suffered a serious data leak: customers' conversations with its corporate AI chatbot, including phone call recordings and text chats, were left publicly accessible on the internet. Anyone could view them without logging in and without any special technical skills.

According to an investigation by Wired, what was exposed was not just technical logs. The conversations contained customers' names, contact details, email addresses and specific purchase details: when the purchase was made, what was bought and what problems came up. This is exactly the kind of information scammers prize: knowing a person's name, what product they bought and when they contacted support, an attacker can build a convincing phishing scenario.

The attack looks like this: a fraudster calls a Sears customer and poses as a company employee. "Hello, Maria. We're calling about your refrigerator order from April 12. There's a problem with the delivery." The person smells nothing suspicious: the details match down to the last item. Next comes a request to confirm payment details, pay for a "repeat delivery" or click a link to "confirm the address." This scenario is called spear phishing. It is far more effective than mass mailings precisely because it uses the victim's real personal data.

The Sears incident is not an isolated case but a symptom of a systemic problem. As retailers, banks and service companies roll out AI chatbots for customer service en masse, the volume of personal data these systems collect and store is growing rapidly. Modern chatbots hold full conversations, remember the context of previous requests and have access to order history. That makes them useful support tools and, at the same time, dangerous points of risk when data storage is misconfigured.

From a security standpoint, chatbot conversation logs are especially valuable to attackers for several reasons. Customers talk to bots candidly, describing problems in detail they would not share with a stranger. The data is structured and easy to process: name, date, type of request make a ready-made profile for an attack. Finally, users do not expect their technical exchange with a bot to become public, and they are not as careful as when filling out official forms.

Sears' response to the incident remained unknown at the time of publication. The company is going through a rough period: once the largest American retailer, it has shrunk to a small chain after years of financial difficulties and bankruptcy in 2018. Adopting AI tools to cut support costs fits logically into a survival strategy, but skimping on data security creates risks that may prove far more expensive.

The case raises a broader question: how ready are companies to deploy AI systems that handle sensitive data responsibly? In the EU, the AI Act, with its risk assessment requirements, is already in force. In the US, the FTC has consistently pursued companies for inadequately protecting consumer data, and leaks of this scale clearly fall within its interests.

For ordinary users, the takeaway is simple: an unexpected call from someone who knows too much about your purchases may not be a coincidence, but a direct result of a leak that nobody warned you about.