RSA warns banks of exploding shadow AI agents

RSA Security’s chief executive Greg Nelson warned that banks could see a five‑fold surge in hidden AI agents—software bots that act autonomously—within a year, a trend that could strain oversight and expose institutions to costly errors.

What the "shadow agent" explosion looks like

Nelson, who became RSA’s CEO in September 2025, told PYMNTS that a midsize bank that believed it had no AI agents was actually running about 4,000. He predicts that number could swell to 20,000 in the next twelve months if banks do not tighten governance.

• Current estimate: 4,000 hidden agents in a typical midsize bank.
• Projected growth: Up to 20,000 agents within 12 months.
• Industry signal: Roughly 40 customer meetings in the past month revealed widespread uncertainty about agent inventories.

These “shadow agents” are AI‑driven programs that continue to operate even after the employee who deployed them leaves the firm, often retaining permissions that exceed their original scope.

Why the risk matters to investors and consumers

For investors, the rapid proliferation of unmanaged agents raises two key concerns. First, regulatory scrutiny could intensify as supervisors demand proof that banks can account for every autonomous process handling high‑value transactions. Second, operational failures—such as an agent mistakenly wiring $50 million—could trigger costly remediation, legal exposure, and reputational damage.

Consumers may feel the impact indirectly. If a bank’s approval queues become overloaded, routine requests (like loan applications or fund transfers) could face delays, while high‑risk decisions receive insufficient review. In worst‑case scenarios, a rogue agent could move funds without proper human sign‑off, eroding trust in digital banking services.

Nelson’s analogy underscores the stakes: “If an agent is looking to wire $50 million, you do not just want that to naturally occur.” The warning aligns with broader industry anxiety, as reflected in Microsoft’s AI chief cautioning about unchecked AI autonomy.

How banks can respond

RSA proposes a three‑step approach:

1. Discovery: Conduct a comprehensive inventory to identify every active AI agent, regardless of its apparent relevance.
2. Accountability: Assign a human owner to each agent, ensuring that the agent’s permissions never exceed those of its sponsor.
3. Governance: Apply tiered oversight—routine agents receive basic ownership, while high‑risk agents (e.g., those capable of moving large sums) trigger explicit human review and tighter controls.

Nelson warns that banning AI tools outright could backfire, pushing employees toward unsanctioned solutions that are even harder to monitor. Instead, banks should focus on “reliably find[ing] and govern[ing] the software that’s already in use,” a strategy that balances innovation with risk management.

What to watch next

Stakeholders should monitor three developments over the coming months:

• Regulatory guidance: Expect clearer expectations from banking regulators on AI agent inventory and permission reporting.
• Industry benchmarks: Look for emerging standards—potentially driven by firms like RSA or consortia such as the Financial Services Information Sharing and Analysis Center (FS‑ISAC)—that define “critical” versus “routine” agents.
• Technology solutions: New AI‑governance platforms, including those from startups like Dataline, may gain traction as banks seek automated discovery and permission‑mapping tools.

By tracking these signals, investors can gauge which banks are proactively managing the shadow‑agent risk and which may face costly surprises.

Sources

• RSA Warns Banks of the Risk of the Shadow Agent Explosion — pymnts.com