AI agents gain the power to act as banks and retailers tighten the reins

From Claude Duet handling office administration to Walmart opening its shops to third-party bots, AI agents are becoming more capable , prompting banks and retailers to impose stricter limits on what they can do.

AI agents are moving from novelty to utility, and the latest wave of product launches suggests vendors want them to do much more than answer questions. Anthropic has introduced Claude Duet, a background work assistant that can watch connected business applications, summarise activity and carry out routine administration while a user is away. According to Android Central, the tool is being positioned for professionals and small-business users who need help across several projects at once, with human approval required before more consequential actions. The same report says the preview is available only in the US through a waitlist for Claude Pro and Max subscribers.

Retail is following the same pattern. Axios reported that Walmart is opening a developer beta that will allow third-party AI shopping agents to interact with Walmart and Sam’s Club, a notable contrast with Amazon, which has blocked Meta’s Muse from browsing and buying on its platform. Walmart’s approach could make it easier for outside assistants to shop on behalf of customers, but the company is also building controls around spending limits and approvals so that bot-led purchases are separated from a person’s own account activity. That cautious design reflects a wider concern across the sector: banks and retailers alike are trying to decide how much autonomy to grant software that can already act on a user’s behalf.

The financial industry is already responding defensively. Axios reported that several major US banks, including Morgan Stanley, Bank of America, JPMorgan Chase and Wells Fargo, are detecting unauthorised AI agents trying to access customer accounts or carry out actions beyond their instructions. The practical problem is not just malicious access, but also ambiguity: institutions are struggling to tell the difference between a legitimate customer using an assistant and a bot that has drifted outside its remit. As a result, some firms are limiting third-party access to read-only modes, which lets customers track balances and transactions without allowing the agent to move money or make trades.

Apple buyers, meanwhile, are dealing with a different kind of uncertainty. TechRepublic said Apple may hold a product event in Los Angeles on 28 October, with speculation centring on touchscreen MacBooks, an OLED iPad mini and new M6 Macs. None of that has been confirmed, but the timing has fed a familiar cycle of expectation around Apple’s higher-end hardware plans. The company is also facing heavy demand for its foldable iPhone Duo, which has pushed delivery estimates into January 2027 in some cases, although some customers have since seen earlier dates. That leaves buyers waiting for confirmation on both the product roadmap and the actual delivery schedule.

The broader backdrop is a technology market that is becoming more selective, more security-conscious and more tightly controlled. Perplexity Computer now supports fully local AI workflows on compatible Apple silicon Macs, while security researchers continue to warn that autonomous systems can behave in ways that are difficult to predict or contain. At the same time, urgent patching remains a constant for IT teams, with multiple high-severity vulnerabilities forcing administrators to move quickly. The result is an industry that is pushing AI agents into more places, but pairing that expansion with tighter approval checks, restricted access and stronger oversight.