Identity Verification Weekly: Rogue AI Agents, Dutch ID Wallets, and Japan’s eKYC Shift

Major US banks, including Morgan Stanley, Bank of America, JPMorgan, and Wells Fargo, are encountering third-party AI agents accessing customer accounts, Axios reports. In some cases, agents use customers' login credentials and access accounts with their permission, even though this may violate banks' rules requiring customers to log in themselves.

Banks have also observed AI agents contacting call centers or apparently attempting actions beyond users' instructions. According to Axios, many financial institutions struggle to distinguish between agents acting on legitimate customer requests and those operating without proper authorization.

Some banks are already taking action. Morgan Stanley has begun blocking third-party agents from its platforms, while JPMorgan, Bank of America, and Citi are restricting access. Meanwhile, the industry is exploring read-only access as a way to let AI agents retrieve account information without allowing them to initiate transactions.

Why it matters: AI agents introduce a new layer of complexity into banking security. Verifying the customer and confirming that an agent can access an account are separate from determining whether it has permission to perform a specific action.

Banks need to connect the verified customer, the instructions they gave, and the transaction an agent is attempting. They also need to determine when an action exceeds the original permission and requires the customer to step back in. Depending on the risk, that could mean confirming a transaction or verifying the customer's identity again.

India’s cybersecurity agency, CERT-In, has reported cases of AI agents accessing confidential information in Indian organizations.

According to the report, the agents exceeded the authorization granted for their tasks and accessed live systems containing business records and personal information. The affected organizations did not detect the activity themselves. Instead, AI model providers identified and voluntarily disclosed the incidents.

CERT-In also reported separate cases in which attackers deliberately used publicly available AI tools to compromise systems for financial gain, accelerating attacks that previously took days into a matter of hours.

Why it matters: The incidents highlight a growing challenge for organizations using autonomous AI: an agent may have legitimate access to a system but still act beyond the task it was assigned. Valid credentials alone cannot establish whether every action an agent takes is permitted.

Organizations need to define what agents can access and do, enforce those limits, and monitor their activity for unexpected behavior. They also need a clear way to trace actions back to the authority granted, so they can investigate incidents and determine where controls failed.

A coalition of Dutch financial, telecom, and consulting companies has unveiled MyCheck, an independent digital identity wallet designed for everyday online interactions.

Developed with the EU’s eIDAS 2.0 framework in mind, the wallet aims to give users greater control over the personal information they share. For example, someone could prove they meet an age requirement without disclosing their full name or date of birth.

MyCheck also aims to simplify access to participating services through a common digital identity experience. Its founders have offered to work with the Dutch government on deployment, while several major organizations have expressed interest in adopting the wallet.

Why it matters: MyCheck is part of a broader shift toward digital identity wallets across Europe. Under the EU's revised eIDAS framework, Member States are required to make EUDI Wallets available by the end of 2026. With adoption expected to expand in 2027, initiatives like MyCheck are likely to become increasingly common as governments and private companies develop new ways to use verified digital credentials.

For businesses, digital wallets could make identity checks faster and more convenient while reducing unnecessary data collection. Instead of repeatedly uploading identity documents, customers could share verified credentials or specific attributes, such as proof that they meet an age requirement. However, businesses will still need to establish that these credentials are authentic, valid, and presented by their rightful holders. Supporting different wallets and ensuring interoperability with existing verification systems will add another layer of complexity.

The UK’s Government Digital Service has opened applications for registered digital verification service (DVS) providers to join the GOV.UK Wallet sandbox.

The development marks a concrete step toward allowing private-sector verification services to test how they can work with government-issued digital credentials, including the planned digital driver’s license.

The government has been discussing industry involvement for some time. In 2025, it outlined plans to engage verification providers, and earlier in 2026 it said they would be able to test digital driver’s license data sharing ahead of a wider rollout. The opening of sandbox applications now gives that work a more practical route forward.

Why it matters: For digital IDs to become useful beyond government services, businesses need practical ways to verify them. Opening the sandbox gives verification providers an opportunity to test how government-issued credentials can be shared, checked, and incorporated into existing customer journeys.

It also brings interoperability questions into focus. Organizations will need to establish whether a credential is genuine and valid, understand which information they can rely on, and handle the data securely. Testing these processes before wider adoption can help identify technical and operational gaps.

In Japan, the payment service PayPay has announced plans to discontinue photo-based identity verification using My Number Cards, driver's licenses, and driving history certificates by early November 2026. Instead, users will verify their identity through methods that read the RFID chips embedded in these documents.

The company cited growing concerns about the misuse of identity document images following cyberattacks and data leaks. The change also comes ahead of new Japanese requirements scheduled for April 2027 that will restrict the use of document images in certain remote identity verification procedures.

PayPay is part of a wider shift. According to The Japan Times, major Japanese banks are also moving toward chip-based checks. Mizuho Bank has already discontinued photo-based verification for online account opening, while other institutions have introduced similar changes or are preparing to do so.

Why it matters: The shift reflects growing concern about relying on images of identity documents for remote verification. Stolen or manipulated document photos can be reused in fraud attempts, while reading data from an authentic RFID chip provides a reliable means of checking the document's integrity and authenticity.

For financial institutions, the transition also has practical implications. Chip-based verification requires compatible devices, suitable verification technology, and processes for customers whose documents or devices do not support the preferred method. Japan's experience shows how security concerns and regulatory requirements can accelerate changes in established KYC (Know Your Customer) practices.