Banks still rely on COBOL, a programming language developed in 1959, to process financial transactions. Now, OpenAI, Anthropic, IBM and Amazon are developing AI tools to help banks modernize these aging systems without disrupting essential services.
COBOL remains a key part of banking infrastructure, supporting systems that handle ATM withdrawals, credit card payments and account balances. However, replacing this software is difficult because decades of business rules are built into millions of lines of code, much of which is poorly documented.
AI coding tools could help banks analyze legacy applications, identify connections between systems and translate old code into modern programming languages. An October 27 report from The Verge examines whether these advances could make COBOL modernization easier.
Early results are promising, but significant challenges remain. Translating code does not guarantee that a replacement banking system will work correctly. Even a small error could affect payments, account balances or transaction records.
Banks must therefore verify that AI-generated code preserves the original system’s behavior before putting it into use.
ALSO READ: AI Coding Tools Statistics, Market Size and Growth 2025-2026
Why Banks Still Depend on a Programming Language From 1959
COBOL stands for Common Business-Oriented Language. It was created in 1959 to help businesses and government agencies manage large amounts of data, such as payroll records, financial transactions and administrative information.
Banks adopted COBOL because it handles financial calculations and processes large numbers of transactions reliably. Many banks built their core systems around the language and continue to rely on them today.
More than 60 years later, COBOL remains an important part of the banking industry. According to figures attributed to IBM and cited by CIO Dive in July 2026, the language still supports a significant share of the world’s banking activity.
| Banking Activity | Commonly Cited COBOL Involvement |
| ATM Transactions | 95% |
| In-person Credit Card Transactions | 80% |
| Online Banking Systems | Over 40% |
Source: CIO Dive
These numbers show why replacing COBOL has become a major challenge for banks. Banks cannot shut down systems that handle millions of transactions and replace them with new software overnight. These systems must remain secure, accurate and available throughout the process.
Many COBOL applications have also been updated repeatedly over several decades. Banks have added new rules to meet changing regulations, introduce new products and connect older systems with newer technologies.
Some of these rules are buried deep within the existing code, making them difficult to identify and understand. Modernizing COBOL systems involves more than converting old code into a new programming language. Banks must also understand how their existing systems work and ensure that the replacement software performs the same tasks correctly.
OpenAI Says Its New AI Model Can Translate COBOL With 98% Accuracy
Advances in AI coding tools are creating new opportunities to modernize older software. OpenAI says its GPT-6.1 Sol model achieved 98% behavioral preservation accuracy in a COBOL modernization test when using higher reasoning effort.
The result, reported by The Verge on October 27, suggests that advanced AI models can translate legacy code while preserving how the original software works.
This is particularly important for banks and other financial institutions. Their systems handle payments, account balances, interest calculations and millions of other transactions. Any change in how these systems process information could lead to costly errors.
Behavioral preservation means the translated software produces the same results as the original system under the conditions being tested. For example, an application calculating interest must follow the same rules for interest rates, rounding and account balances after its code is converted.
If the new software changes even one of these calculations, it could create financial discrepancies despite running without technical errors. OpenAI’s reported 98% accuracy is a promising result, but it needs to be understood in context. The figure comes from a specific evaluation. It does not mean AI can modernize an entire banking system with a guaranteed 98% success rate.
The remaining errors also matter. Even a small number of incorrect results could cause serious problems if they affect critical financial transactions.
AI could help banks speed up COBOL modernization and reduce some of the work involved. However, financial institutions must still test the new software carefully to ensure it produces accurate results and works reliably in real-world conditions.
Anthropic Says Claude Can Help Banks Understand Their Old Software
Anthropic says its Claude coding tools can help banks understand old software and simplify COBOL modernization. One of the biggest challenges is identifying how legacy applications work. Many banking systems contain business rules added over several decades.
Their original developers may have left, and existing documentation may be outdated. Before replacing these systems, engineers must understand how the code works, how data moves between applications and which business rules must remain unchanged.
Claude can help analyze COBOL code, document existing systems, identify connections between applications, generate tests and assist with writing replacement code. This could reduce the time engineers spend examining legacy software.
However, AI-generated explanations and code still require human review. Banks must verify that the tools correctly identify business rules before using them to replace critical financial systems.
Amazon Wants AI to Modernize Entire Banking Applications
Amazon Web Services (AWS) is developing AI tools to help banks replace outdated software. Its AWS Transform service can analyze legacy applications, identify business rules and support code modernization.
On September 29, 2026, AWS announced a partnership with banking technology company Thought Machine to speed up legacy banking system migrations. The companies are combining AWS Transform with Thought Machine’s Vault Forge technology to analyze older mainframe applications and turn existing financial rules into new products for testing.
Thought Machine says this approach could reduce some migration processes from years to days. However, this does not mean an entire banking system can be replaced safely in just a few days.
Banks must still check data accuracy, regulatory requirements, security and connections with other systems before deploying new software.
The modernization process involves three main steps: analyzing existing COBOL code, developing replacement applications and testing the new software against the original system.
AWS also stresses the need for human review before AI-generated code is deployed. Banks must verify that the new applications follow the same business rules and process transactions correctly.
IBM Says Replacing COBOL Is Not the Same as Modernizing a Bank
IBM is also using AI to help businesses modernize COBOL applications. The company has spent decades providing the mainframe systems that banks use to process transactions and manage critical operations.
In 2023, IBM introduced watsonx Code Assistant for Z. The tool helps developers understand legacy applications, convert selected COBOL code into Java and test the results.
However, IBM argues that rewriting COBOL code does not automatically mean banks should replace their mainframes. These systems handle transaction processing, security, data recovery and connections with other business applications.
In February 2026, IBM executive Rob Thomas highlighted the difference between modernizing software and replacing the infrastructure that runs it.
IBM also estimates that around 40% of COBOL workloads run on distributed systems such as Windows and Linux rather than mainframes. This means banks can update their applications without necessarily changing their underlying infrastructure.
IBM has reported improvements from its AI modernization tools. In one customer example, the company said its tools reduced the time needed to identify certain unnecessary COBOL routines by 94%, cutting the task from around eight hours to 30 minutes.
These results apply to a specific customer and do not guarantee similar improvements across all projects.
IBM’s approach shows that banks can use AI to improve legacy software while keeping their existing infrastructure where it still meets their needs.
Why a 2% Error Rate Could Be Dangerous for Banks
AI-assisted COBOL modernization must meet strict accuracy requirements. An incorrect chatbot response may cause inconvenience, but an error in banking software could lead to financial losses, customer disputes and regulatory issues.
Consider a banking application that processes one million transactions. If 2% were processed incorrectly, that would mean 20,000 errors. This is a hypothetical example, not a prediction based on OpenAI’s reported 98% behavioral preservation score.
Banks must test replacement software to ensure it follows the same rules as the original system. Testing should cover unusual transactions, account histories, financial calculations and system failures.
The new software must also work correctly with databases and other connected banking applications. An error in one component could affect several services. Banks can reduce these risks by running old and new systems in parallel and comparing their results before switching over.
They can also use independent code reviews and formal verification to check specific software properties. AWS has also highlighted the importance of giving AI coding tools enough information about existing code, dependencies and business rules. If a model misses connections between applications, it may produce incorrect results.
For banks, successful COBOL modernization requires more than code that runs. The replacement system must produce accurate results and work reliably across the wider banking infrastructure.
ALSO READ: Bank of America Hires 1,000 Contractors to Fix AI-Detected Security Flaws
AI Could Reduce the Cost of Modernizing Banking Systems
Modernizing legacy banking systems can be expensive. Banks often need large teams of engineers, consultants, security experts and business analysts to understand and replace old applications. These projects can take years to complete.
AI coding tools could reduce some of this work by helping developers identify software dependencies, document business rules, generate tests and translate COBOL code. This would allow engineers to spend more time on complex tasks.
AWS says its modernization tools can shorten some migration projects from years to months. Thought Machine has also claimed that its technology can speed up certain stages of banking modernization.
However, there is no reliable industry-wide estimate of how much AI can reduce the total cost of replacing core banking systems. The savings will depend on each bank’s existing technology, documentation and testing requirements.
Banks must also account for the ongoing costs of new systems. Cloud services can offer flexibility but may introduce additional expenses for computing, data transfers, security and maintenance.
The goal of AI-assisted modernization should be to improve system reliability and reduce long-term costs, not simply replace COBOL code with a newer programming language.
Will AI Replace COBOL Programmers?
AI coding tools are changing how developers work, but COBOL programmers remain important to banking system modernization. Many legacy applications contain business rules that are poorly documented.
Experienced COBOL developers understand how these systems process transactions and handle financial operations. AI can help analyze and translate the code, but it cannot replace the need for human knowledge and oversight.
In a March 2026 interview with ITPro, AWS executive Asa Kalavade said experienced COBOL engineers would continue to play an important role as AI takes on more modernization work. Developers may increasingly use AI to analyze code, document business rules and generate replacement applications.
Their work will also involve checking AI-generated results and ensuring new systems operate correctly. AI could reduce the time spent on repetitive coding and analysis tasks, but its overall impact on COBOL jobs remains uncertain.
For banks, one key priority is preserving the knowledge of experienced developers before they retire. AI tools could help document how legacy applications work and make that information easier for future teams to use.
Banks May Not Need to Replace COBOL Completely
Banks may not need to replace every COBOL application to modernize their technology. Many legacy systems still handle financial transactions reliably and can continue to serve their purpose.
Instead, banks can update selected parts of their systems. They might keep existing transaction-processing software while building modern interfaces for mobile banking apps and other services. AI tools can also help document old code, identify unused functions and make software easier to maintain.
Other applications may need to be rewritten or moved to newer infrastructure. The right approach depends on each bank’s technical needs, security requirements and long-term costs. COBOL modernization does not always mean abandoning the language. It can involve improving existing applications, converting selected code or rebuilding entire systems.
The main goal is to make banking software easier to maintain, update and connect with modern technology while keeping financial transactions accurate and secure.
AI Could Help Banks Modernize Decades-Old COBOL Systems
Banks have relied on COBOL for decades, but maintaining these older systems can be expensive and time-consuming. Replacing them also carries risks because they handle critical financial transactions.
AI coding tools could help banks address this challenge. Models from OpenAI, Anthropic, IBM and AWS can help developers analyze legacy code, understand business rules and convert applications into newer programming languages.
However, AI cannot replace the need for careful testing and human oversight. Banks must ensure that modernized systems process transactions accurately, protect customer data and meet regulatory requirements.
AI’s biggest contribution may be helping engineers understand and update decades-old software without rebuilding every system from scratch. The goal is to make banking technology easier to maintain while keeping essential financial services reliable and secure.
