TL;DR
Get your next haul 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 identify all the AI agents operating inside their systems. RSA has announced Agent ID to help institutions discover agents, manage permissions and assign accountable people, but the scale of the reported inventory gaps and the product’s capabilities have not been independently established.
RSA Security is urging banks to identify and govern the AI agents operating inside their systems, after CEO Greg Nelson said customer conversations revealed widespread uncertainty about agent inventories. RSA has announced Agent ID to help institutions discover agents, manage their access and assign human accountability, with particular attention to agents that can carry out consequential financial or operational tasks.
Nelson told PYMNTS CEO Karen Webster that about 40 customer meetings in the preceding 30 days had surfaced uncertainty over how many agents were already in use. He said he had not found a company comfortable that it knew its agent count. Those remarks describe what Nelson heard in customer discussions; they are not a published survey or a verified estimate of the banking sector as a whole.
Webster cited an unnamed midsize bank that told RSA it had no agents, while RSA found 4,000 operating in the institution. The report does not identify the bank or explain how RSA counted the agents, so the example cannot be independently checked from the information provided. Nelson also suggested the institution might have 20,000 agents within a year, but presented that as a possibility, not a forecast backed by disclosed data.
Nelson’s central point was that agent count alone does not establish risk. Banks need to know which agents have consequential permissions, who is responsible for them and whether controls match the work. He said an agent should not have more authority than its accountable human owner. An agent seeking to initiate a large wire transfer, he said, should not be allowed to proceed without appropriate scrutiny.
Why Agent Permissions Matter to Banks
AI agents can perform tasks and access systems on behalf of users, so a bank that cannot connect an agent to an owner or understand its permissions may struggle to apply existing controls. The potential concern is not simply that an institution has many agents; it is that an agent could carry out higher-impact work without suitable approval, monitoring or a clear person responsible for it.
Nelson also warned that a nominal human approval step can fail to provide meaningful oversight if staff are overloaded and approve requests in batches. That raises a practical issue for banks: controls should direct human attention toward higher-risk actions, rather than treating every routine request as equally urgent. The interview does not establish that banks have experienced losses or incidents because of agent activity.
Another exposure can arise when an employee leaves. Nelson said an agent may remain active with its permissions intact if the person responsible for it departs. Connecting agent access to a continuing owner, and removing or reassessing that access when the work ends, is part of the governance problem he described.
Popular right now
Vitamix Propel Series 510 Professional-Grade Blender, 3 Programs, Black
Bose QuietComfort Wireless Noise-Cancelling Headphones, Petal Pink
Philips 2200 Series Fully Automatic Espresso Machine with Classic Milk Frother
As an affiliate, we earn on qualifying purchases.
RSA Extends Identity Controls to Agents
RSA has spent decades working on identity and access management, including tools that help organizations determine who can enter their systems. Nelson, who became CEO in September 2025 after serving as president and chief business officer, is applying that approach to software agents as well as human employees.
RSA has announced Agent ID, which the report describes as a product intended to help organizations find agents, manage their access and establish accountability. Nelson said regulated financial institutions need visibility across cloud applications, on-premises systems, mainframes and isolated environments. The report does not provide technical specifications, independent testing or a detailed account of how Agent ID works across those environments.
Nelson argued that prohibiting AI tools outright could push employees toward applications the institution does not oversee. His preferred approach is to discover tools already in use, assign responsibility and match controls to the agent’s permissions and tasks. That is his assessment of the trade-off; the report does not compare the outcomes of restrictions with those of agent-governance programs.
“I haven’t found one company yet that is comfortable that they know how many agents they have.”
— Greg Nelson, RSA Security CEO, speaking with PYMNTS CEO Karen Webster
How Widespread Are Inventory Gaps?
The report does not provide a representative survey showing how many banks lack agent inventories, or how often agents operate with excessive permissions. The 4,000-agent example concerns an unnamed midsize bank and comes through RSA’s account; the counting method and the types of agents included are not specified.
It is also unclear how many agents in the cited institution perform consequential work, whether any were operating without adequate supervision, or whether the bank had a formal inventory before RSA’s assessment. Nelson’s estimate that the count could reach 20,000 within a year is a possibility, not a confirmed outcome.
Details about Agent ID’s availability, deployment process, technical coverage and performance were not included in the source report. The interview also does not document a specific banking incident caused by an AI agent. The risks discussed are prospective and relate to possible gaps in discovery, authorization and human review.
Banks Must Map Agents and Owners
The next step described by Nelson is for institutions to discover the agents already operating, connect each to a responsible person and review what each can access or do. Banks would then need to apply closer controls to agents involved in higher-impact activity and reassess access when an agent’s work or owner changes.
For RSA, further details about Agent ID would help clarify how the announced product addresses those tasks across different systems. The source report does not give a release schedule or name a specific next product milestone. For banks, the practical measure of progress will be whether they can identify consequential agents, confirm appropriate permissions and show that human approvals receive meaningful review.
Key Questions
What did RSA warn banks about?
RSA CEO Greg Nelson said banks may not know how many AI agents are operating in their systems, or which agents have access to consequential tasks. He urged institutions to identify agents, assign accountable people and match controls to the work.
Did RSA say a bank had 4,000 agents?
The PYMNTS report recounts Nelson citing an unnamed midsize bank that told RSA it had no agents; RSA reportedly found 4,000 operating there. The report does not identify the bank or describe the counting method, and the figure is not independently verified in the material provided.
What is RSA Agent ID?
RSA has announced Agent ID as a tool intended to help institutions discover AI agents, manage their access and establish who is responsible for them. The source report does not provide detailed technical specifications or an availability date.
Does the report show that banks have suffered losses from agent misuse?
No. The report discusses possible risks, including excessive permissions, overloaded reviewers and agents left active after an employee departs. It does not document a specific bank loss or incident caused by an AI agent.
Source: pymnts.com
