AI agents gain access to stores and accounts, but security lags behind

An AI agent is a program that doesn't just answer questions but carries out tasks on its own: searching, logging in and making purchases. This week's stories show that more big players are now opening the door to them. It just isn't happening the same way everywhere.

Stores and insurers let the agents in

Walmart has launched Agent Gateway, which lets approved AI agents search inventory and shop for customers at Walmart and Sam's Club. Agent Gateway is so far only in beta for invited users, and agents are not expected to be able to buy on customers' behalf until Agent Profiles rolls out in November 2026. Amazon is going the opposite way: according to TechCrunch, Amazon blocks agent purchases.

In the US, the insurer Kin Insurance has opened its website to AI agents, so paying ChatGPT users can get a home insurance quote. The purchase still takes place with Kin, and the agents cannot assess gaps in coverage.

At Cisco, AI agents can be invited into meetings and rooms in the new Webex as part of the team. Several of the features, including Claude in Webex, are only planned, however, and have no firm date. Oracle has introduced a tool that lets companies build and manage agents with little or no code. It is expected to be available in the US within 12 months.

Access requires control

The more agents are allowed to do, the more important security becomes. 1Password has launched a solution in which agents can log in to a user's accounts without seeing the password itself. The user must approve each request.

Other stories show why this matters. The security firm Forever Security warns of an attack in which a website places a hidden instruction file in the Downloads folder, which AI coding agents then follow. The success rates come from the firm's own tests. Microsoft wants to use so-called Guardian Agents to monitor other agents in companies. The company believes agents often have more permissions than necessary and that the models cannot be blindly trusted.

Privacy is also at stake. The security firm Claroc says it has found more than 6,800 addresses to shared conversations in Meta's AI service Muse, which people can find without having been given the link. Some of the pages contain personal information and internal business documents.

The pace is picking up

Anthropic has launched Claude Duet, an app with an agent that checks your apps, notifies you when something needs your attention, and keeps working while you are away. For now it is only available to Pro and Max subscribers in the US. In Anthropic's own report on the company's employees, agents now work more hours than the staff, and the share of juniors in technical roles has fallen. The figures have not been independently verified and apply to only one company.

What can you do?

If you see an agent asking for access to an account or a payment, read the approval carefully before you say yes. Don't let an agent be the only one assessing, for example, an insurance policy's coverage. Also avoid sharing conversations with AI services if they contain anything you don't want others to see.

Also keep an eye on whether Walmart's Agent Profiles arrives in November 2026 as planned, and whether Amazon keeps its doors closed.