TL;DR
Get smart everyday buys delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
RSA Security CEO Greg Nelson says banks are struggling to inventory AI agents already operating in their systems, citing one midsize bank where RSA found 4,000 after the bank reported none. RSA has announced Agent ID to help organizations discover agents, manage permissions and assign accountability; the report does not establish that those agents were all unsupervised or handling high-risk work.
RSA Security CEO Greg Nelson is warning banks that they may not know how many AI agents are operating inside their systems, as the company promotes Agent ID, a product announced to help find agents, manage their access and assign human responsibility. In an interview with PYMNTS CEO Karen Webster, Nelson cited a midsize bank that told RSA it had no agents; RSA then found 4,000 operating agents, according to the report. The example highlights an inventory and oversight problem, but does not show that all 4,000 had high-risk permissions or lacked supervision.
Nelson said discussions with about 40 customers over the preceding 30 days had surfaced uncertainty about how many agents were already in use. He said he had not found a company comfortable that it knew its full agent count. Those comments describe what RSA heard in customer meetings; they are not an independently measured estimate of agent use across the banking sector.
Nelson argued that the count alone is less important than identifying agents with consequential permissions and checking whether appropriate controls cover them. He said each agent should have a human owner, should not have more authority than that person, and should receive oversight suited to its work. A hypothetical agent preparing a $50 million wire, he said, should not be able to complete that action without suitable review.
RSA has announced Agent ID to help organizations discover agents, manage their access and establish accountability. Nelson said the work must cover agents operating across cloud applications, on-premises systems, mainframes and isolated environments. The report does not provide pricing, a release schedule or independent product testing.
Why Agent Permissions Matter to Banks
A bank cannot reliably oversee an agent if it does not know the agent exists, who is responsible for it or what systems it can reach. That can make an incomplete inventory a governance and security concern, particularly where software can act on financial processes or sensitive data. The report, however, does not establish that the cited bank suffered a breach, loss or unauthorized transaction.
Nelson’s emphasis on permissions points to a distinction between ordinary automation and agents able to take consequential actions. A large number of low-risk agents may call for a different response from a smaller group with authority to move funds or alter important records. Banks therefore need controls that reflect what agents can do, not simply a total count.
Human approval also has limits if employees approve requests in batches without examining them. Nelson said staff can face dozens of requests and pressure to move work along. A review step may exist on paper while providing little scrutiny in practice. Risk-based review can reserve human attention for decisions where it matters most, though the report does not detail a specific control standard or regulatory finding.
Products Worth Considering
Canon SELPHY CP1500 Compact Photo Printer (White)
Dell UltraSharp 27 Thunderbolt Hub Monitor (U2726DER)
FLEXISPOT 60 × 24-Inch Electric Standing Desk, White
As an affiliate, we earn on qualifying purchases.
From Employee Access to Agent Identity
RSA has worked on identity and access management for more than 40 years, including tools used to verify employee access to organizational systems. Nelson became CEO in September 2025, after serving as president and chief business officer. The agent initiative extends that access-management focus from human users to software that can perform tasks on their behalf.
The issue is not only how an agent is introduced. Nelson said an agent may remain active after the employee who set it up leaves, potentially keeping permissions unless the organization reviews and revokes them. He said businesses have paid more attention to starting agents than retiring them. That makes ownership changes and access removal part of the same governance process as initial approval.
Nelson cautioned against responding to the problem by simply banning tools, saying employees could turn to applications the bank does not oversee. His proposed direction is to discover agents already in use and govern them according to their permissions and activity. RSA’s product announcement reflects that approach, while the report does not establish how broadly banks have adopted Agent ID.
“I haven’t found one company yet that is comfortable that they know how many agents they have.”
— Greg Nelson, RSA Security CEO, speaking to PYMNTS CEO Karen Webster
What the Agent Count Does Not Show
The report does not identify the midsize bank or explain how RSA counted its 4,000 agents, what tasks they performed, or how many had access to sensitive systems. It also does not say whether those agents were unauthorized, unowned or operating without controls. A discrepancy between an institution’s estimate and a vendor’s discovery results indicates an inventory gap, but does not by itself prove a security incident.
RSA’s customer-meeting figures are based on Nelson’s account. The report gives no methodology for the roughly 40 meetings, no representative sample of banks and no independent sector-wide data. Nelson suggested the cited bank might have 20,000 agents within a year, but that was a projection, not a verified forecast.
Details about Agent ID also remain limited in the report. It does not specify availability, deployment requirements, pricing, measured results or how the product handles particular regulatory obligations. The effectiveness of any approach will depend on whether institutions can discover agents across their environments and apply controls that match each agent’s actual capabilities.
Banks Must Map and Govern Agents
The immediate step described by Nelson is to discover agents already in use, connect each one to a responsible person and review its permissions. Banks would then need to match oversight to the work performed and remove or reassess access when an agent’s purpose ends or its owner leaves. The report does not name a regulatory deadline or an upcoming enforcement action tied to this warning.
RSA’s next relevant milestone is practical information about Agent ID, including its availability and how it identifies agents across different systems. For banks considering the approach, product details and evidence of how discovery and access controls perform will matter more than an agent count on its own. No adoption figures or customer outcomes for Agent ID are provided in the source report.
Key Questions
What did RSA say it found at the midsize bank?
According to the PYMNTS report, the bank told RSA it had no agents, but RSA found 4,000 operating agents. The report does not identify the bank or explain the agents’ roles and permissions.
Does the report say those agents were unsupervised?
No. It reports a gap between the bank’s stated count and RSA’s discovery, but does not establish that all 4,000 agents lacked owners, had risky permissions or were operating without controls.
What is RSA Agent ID intended to do?
RSA says Agent ID is intended to help organizations find agents, manage their access and establish accountability. The report does not give a release schedule, pricing or independent performance results.
What controls does Nelson recommend?
Nelson says every agent should have a responsible person, should not have more authority than that person, and should receive oversight based on its work. He also says organizations should review and remove access when an agent or its owner is no longer needed.
Source: pymnts.com
