Why Financial Institutions Are Blocking AI Agents and What It Means for Enterprise Identity

In the third quarter of 2026, major financial institutions including JPMorgan, Bank of America, and Wells Fargo began restricting AI agents from accessing customer accounts, marking a critical turning point in how enterprises handle autonomous software. As AI agents transition from passively reading the web to actively executing transactions, legacy identity and access management (IAM) systems are failing to distinguish between legitimate user proxies and compromised bots, forcing a fundamental redesign of digital trust.

How AI Agents Break Traditional Identity and Access Management (IAM) Systems

AI agents act as intermediaries between a human user and a website, obscuring the actual user’s identity from the site operator. Historically, websites could simply block intermediary traffic, such as VPNs or cloud browsers, and require users to disable them. However, as agentic traffic grows to eclipse traditional human browsing, blocking is no longer a viable business strategy.

This creates a severe security blind spot. According to Arkose Labs, “97.7% of legitimate AI agent sessions carried no verified disclosure” of their automated nature in Q3 2026. Furthermore, 65.2% of these agents appeared on login and registration flows, directly challenging traditional access controls.

Because agents do not tire and can operate in parallel, they bypass human-centric friction. As Sardine notes, the days of CAPTCHA as an effective shield are numbered, as large language models are now capable of autonomous computer use and can easily solve visual puzzles.

This dynamic forces a reckoning for site operators. As Cequence Security illustrates, it is akin to a warehouse retailer being told corporate approved thousands of robot shopping assistants, and security must figure out how to stop them from exploiting the system without blocking legitimate members. As Ram Varadarajan, CEO of Acalvio, explained, “an agent holding a customer’s credentials looks identical to the customer”, and a capable one can be hijacked, misled, or even learn to hide what it’s doing, so the bank loses its ability to tell intent from impersonation.

What the OpenID Foundation and PingIdentity Propose for Agentic Identity

To resolve this identity crisis, the OpenID Foundation released a whitepaper addressing how to securely authenticate and authorize autonomous systems. The core challenge is determining whether an AI agent should possess a human-like identity, a strictly non-human identity, or a hybrid. As Aembit explains, AI requires a combination of both, as it acts on behalf of a human but executes like a software workload.

Establishing this requires what PingIdentity defines as digital trust: the confidence that a non-human entity is properly identified, authorized to perform only permitted actions, and that a real person remains accountable. This shifts the focus from simple access control to continuous runtime identity verification.

"The business relationship between Users and Sites, and each participant’s security responsibilities, must be upheld regardless of intermediation." — Cloudflare

This Intermediation Principle dictates that sites need new tools to recognize, serve, and hold accountable the users behind the agent. Reco.ai adds that managing these Non-Human Identities (NHI) is uniquely difficult because agents combine human intent with machine execution, requiring governance frameworks that track both the user’s original goal and the agent’s autonomous path to achieve it.

Why the AI Agent Identity Management Market Will Reach $7.8 Billion by 2035

The financial and operational stakes of this identity crisis are driving massive market growth. The AI Agent Identity Management Market is projected to hit $7.8 billion by 2035, fueled by the sheer volume of non-human identities entering the enterprise. Enterprises are adding AI agents, applications, and workloads far faster than they add employees, meaning the non-human population will soon be several times larger than the human one.

This shift happened because AI agents recently moved from reading the web to transacting on it. The institutional line in the sand drawn by major banks highlights the immediate risk: without proper identity governance, autonomous agents can execute unauthorized financial transactions.

However, businesses face a competitive dilemma in how they respond. Chris Hughes, vice president of security strategy at Noma Security, noted that institutions “don’t want to fall behind competitively by shutting out AI and agents”, because customers are adopting them quickly and a bank that blocks everything risks becoming the less useful option. This tension between strict security and customer convenience is the primary driver for the rapid expansion of agentic IAM platforms.

What IT Decision-Makers Must Do to Secure Non-Human Identities with ManageEngine

For IT leaders and business owners, the immediate priority is establishing strict AI agent accountability. ManageEngine SecureAI outlines that accountability comes down to answering three specific questions: who owns the agent, what is it authorized to do, and what has it actually done.

To achieve this, organizations must move beyond static credentials and implement ephemeral, certificate-based access for agents. Every agent requires a verifiable identity and a narrow set of permissions before it touches a production system. Developers must stop treating agent credentials as a simple secrets database and start building dedicated workload IAM systems that enforce just-in-time access controls.

Furthermore, businesses must deploy continuous behavioral monitoring to distinguish between the 77.4% of AI agent sessions classified as legitimate and the adversarial bots that mimic them. The era of treating all automated web traffic as malicious bots is over. Enterprises that fail to build dedicated identity governance for AI agents will either lock out their own customers’ automated assistants or leave their systems vulnerable to hijacked autonomous workflows.