AI Can Learn COBOL in a Day — But Should It Interact with Financial Infrastructure?

Introduction

Recent demonstrations have shown that advanced artificial intelligence models can understand and generate legacy programming languages such as COBOL in remarkably short timeframes.

This signals how rapidly AI capability is evolving.

But in financial services, the central challenge is no longer whether AI can learn.

The challenge is control.

Financial infrastructure is not experimental software. It underpins market stability, liquidity management, capital allocation, and regulatory accountability.

As AI systems become increasingly capable of interpreting complex architectures, institutions must carefully evaluate whether such systems should interact directly with critical financial platforms.

Financial Infrastructure Is Not Ordinary Software

Core financial platforms — including derivatives, risk, and settlement systems — manage:

• Trade lifecycle processing

• Margin and collateral calculations

• Counterparty exposure monitoring

• Liquidity management

• Regulatory reporting

These systems are:

• Deeply integrated

• Highly regulated

• Interconnected across institutions

• Operationally sensitive

Unlike consumer applications, errors in financial infrastructure can propagate across counterparties, clearing networks, and markets.

The consequences extend beyond a single system.

Capability vs. Accountability

AI systems can now:

• Generate and review code

• Analyze system documentation

• Suggest configuration logic

• Automate workflow scripts

These capabilities offer operational efficiency.

However, direct interaction with production financial systems introduces serious governance considerations.

Before allowing AI to:

• Write scripts

• Modify configurations

• Automate workflows

• Interact with live production data

Institutions must address fundamental questions:

• How is AI-generated logic validated?

• What change management frameworks apply?

• Who bears accountability for configuration errors?

• How are model risk governance policies extended to AI systems?

In regulated environments, oversight is structural — not optional.

Governance Before Autonomy

Traditional financial models operate within formal model risk management frameworks. They require:

• Independent validation

• Ongoing performance monitoring

• Clear documentation

• Defined accountability structures

AI systems introduce additional complexity:

• Probabilistic outputs

• Dynamic learning behavior

• Opaque internal reasoning

• Dependency on large-scale data inputs

Therefore, AI integration must occur within:

• Controlled environments

• Segregated development and production systems

• Human-in-the-loop approval workflows

• Robust audit trails

AI may assist analysis.

It should not operate autonomously within critical infrastructure without structured governance.

The Systemic Dimension

Financial markets operate as interconnected networks.

If AI-generated configuration logic were to introduce errors across multiple institutions simultaneously, the effects could extend beyond individual firms.

AI integration is therefore not solely an operational decision.

It is a stability consideration.

Prudent adoption requires:

• Cross-functional oversight

• Risk and compliance alignment

• Cybersecurity boundary enforcement

• Scenario testing and stress simulation

Innovation must be engineered responsibly.

Conclusion

AI learning legacy languages in days is impressive.

But financial infrastructure is not a coding experiment.

It is a stability mechanism for markets.

The future of AI in finance will not be defined by how intelligent models become — but by how responsibly institutions integrate them into regulated environments.

AI should be the co-pilot. Not the captain. Not yet.