As personal AI agents aimed at individual users spread, whether to open access to these agents has emerged as an important agenda for website operators. For now, more companies across finance and consumer goods appear to be locking their doors to AI agents.

AI agents that carry out everyday tasks on behalf of users may be an opportunity for consumers and developers. But some say they could also burden companies that run websites in various ways.

In finance, major banks including Citigroup, Bank of America (BofA), JPMorgan and Charles Schwab have imposed blocking measures against external AI agents. The banks cited security concerns, but an analysis says they also considered the possibility that large sums could flow out of bank deposits if AI agents maximise returns on behalf of consumers. Banks are raising the issue with regulators, saying deposits could be withdrawn as AI agents move funds elsewhere in search of higher yields, the Financial Times (FT) reported.

Large financial companies are successively blocking access by personal AI agents. "Concern over fund outflows."

Beyond banks, various consumer-facing companies are also putting up barriers against AI agents. According to The Verge, review platform Yelp said it would allow access only to AI agents that pay for data licences, and Delta Air Lines is blocking flight bookings made on behalf of users.

AI agents are also a dilemma in e-commerce. A recent Wall Street Journal (WSJ) report, citing data from search engine and AI search optimisation firm BrightEdge, showed agent-driven purchases have risen about fourfold since April and are growing 43 percent per month. Even so, only 3 percent of e-commerce companies working with BrightEdge allow agents to execute transactions on behalf of consumers.

An analysis says the moves reflect a lack of clear answers on how to respond to a surge of AI agents, even as firms have built capabilities in search-traffic analysis and social media use. Actual consumer mouse clicks and searches are easy to track, but external AI agents often do not clearly indicate whom they represent, making it difficult to grasp what customers want now and in the future, the WSJ reported.

Marketers also need to consider that different AI models evaluate purchase candidates differently. Jim Yu (짐 유), BrightEdge's chief executive officer, said, "This complicates marketers' work, but it's an interesting situation."

Allowing AI agents access could also come with significant costs. In addition to weakening market dominance and shaking customer relationships, mistakes in the purchasing process could increase product returns or credit card payment cancellations. Christopher Waller (크리스토퍼 월러), a member of the U.S. Federal Reserve Board, said, "When AI agents intervene, it becomes much more complex to verify the identity of transaction parties and validate the legitimacy of transactions." He added, "The question is shifting from confirming that the buyer is someone with legitimate payment authority to proving that the AI agent has authority to pay on the buyer's behalf."

For big e-commerce platform companies, the fact that they may lose more than they gain by allowing AI agents to transact could also influence their AI agent policies.

Amazon has blocked AI agents such as Meta Muse and OpenAI Dots from shopping on its platform. With many Amazon Prime subscribers paying for subscriptions, they are unlikely to switch shopping platforms just because access by AI agents is blocked, the WSJ reported. Scott Devitt (스콧 데빗), an analyst at Rosenblatt Securities, said in a recent report, "Amazon, which has a platform where everything from product search and transactions to order fulfilment and data happens within a single ecosystem, has little reason to hand over even part of its value chain to external AI agents."

Retailers with less to lose than Amazon are actively embracing AI agents. Target has opened access to approved external corporate AI agents, and Walmart is also working with OpenAI and Google to increase traffic through AI tools.

E-commerce companies are also accelerating the development of their own AI agents. Delivery platform DoorDash unveiled its own agent, Ask. Customers may trust recommendations from external AI agents more than companies' own agents, but agents built directly by companies can understand inventory and promotion information far more accurately and respond more easily to customer issues, the WSJ reported.

Regardless of whether companies block AI agents, bot-blocking systems used to prevent fraud and hacking are also obstacles for AI agent companies such as Meta. Cloud and security firm Cloudflare introduced a one-click blocking feature for AI crawlers in 2025, and it also supports blocking crawlers that scrape both search indexing and AI training data.

On Reddit, a post said Google blocked some accounts using Muse for "suspicious activity", and cases have also emerged at Walmart, which is partnered with Meta, where payments did not go through. Such situations could lead more websites to strengthen procedures to verify whether users are human, making web use more cumbersome, The Verge reported. Chris Cannell (크리스 캐널), chief executive officer of AI safety evaluation firm EquiStamp, said, "If companies actively block AI agent activity, there will be fewer useful tasks that can be handled online."