Banks' AI Push Runs Into Three Fears About Employee Change

By Elena Vasquez

Three fears dominate bank employees' surveys about AI: losing their job, distrusting the technology, and having no time to learn it, according to Vikram Sivakumar.

The fears sit beside cyber threats, fraud, third- and fourth-party exposure, and operational risks from customer-facing or bot-facing models, Kirti Choudhary said.

Sivakumar described human change as the crux of the challenge, while Choudhary put culture at the center of adoption. Their exchange located AI's hardest work in people and governance, not just in tools.

What risks do banks need to control?

Interviewer Chana Schoenberger opened by asking where the sector's biggest AI risks sit today. Sivakumar began with a premise close to banking itself: banks manage risk by nature.

For Sivakumar, maintaining the bank's brand and customer trust is paramount. Competitive pressure changes how banks manage that duty, because waiting also carries risk.

Sivakumar put his approach this way:

“When you talk about AI, it's not about how do I operate at speed and safely, it's about how do I operate safely and at speed in that order.”

His answer started with the bank's existing obligation to customers rather than a new technical danger. It also assumed that banks can adapt their established risk discipline to AI.

Sivakumar argued that banks already have years of experience balancing risk. AI adds a new muscle to develop, but inaction remains a competitive hazard.

Choudhary broadly agreed that risk management comes naturally to banks. Choudhary also asked how institutions can avoid becoming too conservative while anticipating different problems.

The central standard was behavioral. Choudhary asked how banks can avoid treating AI and its models as if they should sit above the way a human or employee behaves in the same setting.

Choudhary named four risk priorities:

• Cyber threats.
• Fraud.
• Increasingly third- and fourth-party risks.
• Operational risks when models become customer-facing or bot-facing.

Together, those categories connect model behavior to the bank's wider accountability. Choudhary's concern was that new systems could create exposure where banks have traditionally expected clear rules and human judgment.

Where does responsibility sit when two agents meet?

Schoenberger then asked what happens when a customer's agent talks to a bank's agent, while a human exists somewhere but remains largely uninvolved. Her question made the control gap concrete.

Choudhary called that a large liability question. The issue extends beyond whether each model performs its assigned task. It reaches the bank when automated systems interact and human oversight stays thin.

Choudhary proposed a pragmatic sequence: run experiments, observe what happens, learn from the results and make progress. The approach avoids waiting for a complete answer to every edge case.

This gives banks a route forward before certainty arrives. It also answers Sivakumar's competitive concern, since waiting can become a risk when other institutions continue moving.

The discussion did not settle the accountability question. It treated experiments as a way to learn while the larger issue of responsibility remained open.

Why do the same three employee fears keep returning?

Sivakumar said deploying tools and completing organizational change are separate jobs. Banks are beginning to recognize the change-management problem, but a complete playbook is not yet available to every organization.

In the work Sivakumar described, teams begin with an employee survey. The same three concerns repeatedly emerge, although their order changes.

“I'm afraid of losing my job. I don't trust AI and I don't have time to learn.”

Sivakumar's response combined three elements: a structured approach, an environment where teams can learn, and a realistic path for behaving differently at work.

He also said leaders must show employees that they will have jobs after the transformation. The point is not to remove jobs, but to give workers a credible place in the changed organization.

That makes workforce planning part of AI risk management rather than a communications exercise after deployment. Sivakumar said banks need a more mature and explicit conversation about reskilling and responsible treatment of talent.

What must leaders do beyond deploying AI tools?

Sivakumar said leaders must model the change instead of mandating it. He described town halls where employees can discuss their favorite use cases with authenticity and detail.

The leadership task extends from the front line through middle management to the top. Each group has to help shape the new way of working, rather than receive a finished policy.

Boards are increasingly asking organizations about their pathway into a post-AI workforce. Sivakumar said those questions include how banks will care for talent and build upskilling and reskilling paths.

The board-level question rejects the idea that AI automatically means fewer people are needed. It asks how institutions can prepare employees for roles that do not yet have settled definitions.

Choudhary added culture to that agenda. Leaders must signal that they want AI-first institutions while making clear that progress should not come at humans' expense.

For Choudhary, that means bringing employees along during the transition. Workers should see themselves as augmented rather than replaced, and leaders should acknowledge that new jobs will be needed in the AI era.

Can experiments move AI faster than certainty?

Choudhary's experimental approach offers a practical path for banks. Institutions can test uncertain interactions, learn from failures and widen deployment only as their controls improve.

Those tests still have to operate inside the risk framework Choudhary described. Banks cannot treat cyber threats, fraud, outside-party exposure and customer-facing model behavior as separate from the deployment decision.

The human measures matter just as much. Structured learning, visible career paths and leadership willing to model the change address the three fears that Sivakumar's surveys repeatedly uncover.

The next proof point for bank AI will be organizational rather than technical. Employees need credible roles after transformation, and managers need tests that clarify accountability when agents interact.

That is the adoption threshold Sivakumar and Choudhary described. AI can move faster through controlled experiments, but banks must earn trust while doing it.