Why UK Banks Can't Afford to Wait on AI-Driven Mainframe Security

American banks are putting AI to work on their mainframes. UK banks are still asking permission.

That gap won't close on its own, and the cost of waiting rises every quarter. US institutions are piloting AI-driven threat detection directly on their core systems. Many UK banks remain in a familiar pattern: careful committees, pilots that never leave the lab, and a belief that the mainframe's reputation for resilience protects it from modern risk.

That belief no longer holds. Soon, the greater risk may be doing nothing.

AI is speeding up how fast attackers find and exploit vulnerabilities. Security teams will face a volume and pace of remediation work they were never staffed or structured to handle manually. Banks that stick with yesterday's operating model aren't just delaying AI adoption. They're letting a gap widen between how fast threats evolve and how fast their teams can respond.
 

The mainframe isn't isolated anymore, and attackers know it

For decades, the mainframe earned its place as the most secure platform in financial services. That reputation created a risky assumption: if it's isolated, it's safe.

Today's mainframe isn't isolated. It connects to cloud analytics, modern APIs, and a growing web of third-party integrations. Each connection is a potential entry point.

AI has changed what attackers can do with that entry point. Skilled threat actors use AI to speed up vulnerability discovery, build highly targeted social engineering campaigns, and find obscure weaknesses across connected systems faster than human teams can.

That creates the next challenge. More discoveries mean more work. Your team has to evaluate, prioritize, patch, test, document, and monitor every vulnerability. It has to investigate every suspicious event. And it has to complete every fix without putting critical banking workloads at risk.

The security problem isn't just detection anymore. It's execution at scale.

Most mainframe security teams were built for a different environment. Experienced professionals work through queues of alerts, vulnerability reports, access reviews, patches, change requests, and compliance tasks. That model can't scale in a straight line against threats that grow exponentially.
Your team wasn't built for the pace ahead. That's not a failing. It's a signal to change the operating model.

That's where agentic AI comes in. Instead of handing another alert to an already stretched team, governed agents can help investigate, prioritize, and recommend the work needed to fix a problem. Where your policies authorize it, they can help carry out approved actions, with your people in control.
The evidence is already clear. Every vulnerability assessment run with Rocket^® z/Assure^™ VAP has uncovered issues that AI-enabled attackers could exploit. High-severity risks show up in nearly every environment assessed. And 91% of mainframe organizations have experienced a security compromise in the past five years.

The threat is here. The question is whether the operating model defending your mainframe can keep pace with what comes next.
 

The next security gap is a capacity gap

For UK banks, this risk is easy to underestimate.

Picture AI helping your organization find vulnerabilities faster than your team can fix them. You might discover hundreds of issues across applications, middleware, configurations, security controls, APIs, identities, and infrastructure.

Knowing those issues exist is valuable. But knowledge alone doesn't reduce risk. Someone still has to decide which vulnerability creates the greatest exposure, understand its dependencies, apply the right fix, validate the change, and keep an audit trail.

Multiply that across a connected mainframe estate, and the challenge becomes one of capacity.

You can't hire your way out of an exponential workload.

Specialized mainframe expertise is already scarce. Asking those same experts to manually absorb an AI-accelerated stream of findings, attacks, investigations, patches, access reviews, and regulatory demands isn't a sustainable strategy.

That's why agentic AI is more than another security tool. It's a scalable operating model for the work itself.

With the right governance, authorization, and human oversight, agents can:

• Take on repeatable security tasks
• Correlate information across systems
• Speed up investigations
• Recommend remediation steps
• Orchestrate approved workflows
• Keep evidence of every action taken

Rocket^® EVA.Next^™ is built for exactly this. Orchestrated specialized agents work with real-time mainframe context to help your system operators, security, and networking teams move from finding to approved action faster. EVA.Next analyzes data on-premises and works within your existing RACF, ACF2, and Top Secret controls. It's agentic, not autonomous.

The goal isn't to remove people from mainframe security. It's to make scarce expertise go further, so your team can meet a threat environment that's moving faster than any human team can alone.
 

US banks are moving, and the payoff is already visible

Across the Atlantic, the thinking is different. US financial institutions treat AI-driven security as a competitive necessity, not a future project. They're using automation and AI to pursue faster incident response, stronger fraud detection, and more continuous compliance.

The difference isn't access to better technology. UK banks can use many of the same capabilities. The difference is posture.

Leading institutions apply AI and automation to anomaly detection, privileged account monitoring, identity and access governance, and evidence collection across complex environments.

The next step is moving from AI that tells security teams what to do to governed agentic AI that helps them get it done.

That distinction matters more as the workload grows. Another alert isn't transformation. A vulnerability that sits in a remediation queue for six weeks doesn't make you more resilient. The advantage comes from shrinking the distance between detection, decision, and action.

UK banks that wait for a regulatory mandate, perfect certainty, or a peer to move first risk falling behind just as response speed becomes a security capability in its own right.

What's actually holding UK banks back?

Isn't the mainframe already the most secure platform?

It's still one of the most resilient and secure enterprise platforms available. But strong foundations don't justify static defenses.

Your mainframe now connects to APIs, distributed systems, cloud services, identity platforms, middleware, third-party applications, and AI-enabled workflows. As that surrounding attack surface changes, your security operations need to change with it.

AI also changes the economics of attack. When adversaries can investigate weaknesses faster and at greater scale, defenders can't rely on processes built around mostly manual investigation and remediation.

Doesn't adopting AI security tools mean ripping out existing infrastructure?

No. Modernizing mainframe security doesn't require a rip-and-replace overhaul of your core systems.

It means applying AI-driven detection, intelligence, governance, automation, and agentic capabilities to the infrastructure you already rely on.

The goal isn't autonomous change without controls. It's governed action. Agents work within clear permissions, policies, approvals, and audit boundaries. With EVA.Next, PlanGuard evaluates each proposed action when it's invoked and enforces your existing RACF, ACF2, and Top Secret controls. Your people approve what happens next. You speed up security work without giving up the controls your mainframe requires.

Why can't we just add more people?

Because this isn't a headcount problem.

Experienced mainframe security talent is already scarce, and AI will likely increase the security work that needs attention:

• More discoveries create more investigations.
• More findings create more remediation decisions.
• More connectivity creates more dependencies to understand.
• More regulatory scrutiny creates more evidence to maintain.

Specialist headcount can't grow at the same rate as AI-powered threats and vulnerability discovery. The answer is to give the experts you already have far greater leverage.

Why does continuous evidence matter more than point-in-time audits?

Because resilience depends on knowing your controls work all the time, not just when an audit arrives.

Agentic AI takes that further. Beyond collecting evidence, governed agents can help identify gaps, start approved remediation processes, document actions, validate outcomes, and maintain a continuous record of what changed and why. Integrations with tools like ServiceNow and Jira keep that work inside the workflows your teams already use.

That turns security and compliance from periodic exercises into an ongoing operational capability.

The cost of standing still

The real risk for UK banks isn't just adopting AI later than US peers.

It's that the work required to defend the mainframe grows faster than your ability to do it.

Imagine vulnerability discovery increasing tenfold while remediation capacity stays flat. Seeing 10 times as many problems doesn't make you safer. You're safer only when you can prioritize, investigate, fix, validate, and document them fast enough to reduce your exposure.

That's the scalability problem agentic AI can help solve.

The banks that succeed won't be the ones with the most alerts, dashboards, or AI-generated recommendations. They'll be the ones that safely turn intelligence into action at machine-assisted speed.

This isn't about replacing skilled mainframe teams. It's about keeping them from becoming the bottleneck between a faster threat environment and the systems they protect.

Rocket Software builds its approach to mainframe AI around that reality. We help you bring intelligence and automation closer to your most critical systems while keeping the governance, authorization, auditability, and human oversight those systems demand. Find more with z/Assure VAP. Connect more and act faster with EVA.Next. Stay in control every step of the way.

Stop watching. Start acting.

The AI security race isn't about who adopts the most AI. It's about who can respond at the speed AI now demands.

Attackers will use AI to move faster. Vulnerability discovery will accelerate. Your security team will have more signals to investigate, more weaknesses to fix, more patches and changes to evaluate, and more evidence to maintain.

Today's mainframe workforce wasn't designed to absorb that workload manually. And adding people won't scale fast enough.

That's why agentic AI is becoming the critical path forward. Used responsibly, with strong governance and human oversight, it gives your mainframe security team the leverage to keep pace. AI stops being something that only identifies risk and becomes something that helps you steadily reduce it.

UK banks have a window to build that operating model now, before the workload outpaces them.

The question isn't whether AI-driven threats will reach the mainframe. It's whether your security team will be ready when they do.

Talk to Rocket Software about what governed, agentic AI for mainframe security can look like for your institution. We'll help you prepare your security operations for the pace of work ahead, one step at a time.