How the Models Translate Legacy Code
A new update from OpenAI claims its language models can now better assist banks in converting decades‑old COBOL code into modern systems. The announcement came during a press briefing in San Francisco, where the company highlighted improved capabilities for legacy languages such as Fortran and Visual Basic, in addition to COBOL.
Breaking news
tsrs: A Rust Port of the TypeScript Compiler
ByteDance, Alibaba and Tencent Place Massive Order for Nvidia GPUs
Salesforce to Acquire Braze in $5.7 Billion Deal
Eka Robotics Secures $150M at $1.45B ValuationThe update addresses a growing challenge in the financial sector: many banks still rely on COBOL programs written by a shrinking pool of veteran programmers. As those experts retire, institutions risk losing critical knowledge needed to maintain or upgrade core banking functions. OpenAI says its models can read, understand, and rewrite legacy code, reducing the need for specialized legacy developers and accelerating modernization projects.
What This Means for the Banking Sector
OpenAI’s new approach uses a fine‑tuned version of its GPT architecture, trained on a diverse corpus of legacy source files and documentation. According to the company, the model can parse COBOL syntax, identify business logic, and generate equivalent code in languages like Java or Python. In a live demo, the model converted a simple COBOL routine that processed loan applications into a Python script within minutes. The output passed basic unit tests, suggesting the translation preserves functional intent.
Banking analysts note that the ability to automatically generate test‑driven code is a significant advantage. „Legacy systems often lack modern testing frameworks, making migration risky,” said Maria Chen, a senior analyst at FinTech Insights. „If a model can produce code that includes unit tests, it lowers the barrier for banks to validate new implementations.”
Will Banks Adopt AI‑Powered Migration?
The announcement arrives as several major banks announce plans to retire COBOL infrastructure over the next decade. The U. S. Federal Reserve estimates that more than 90 percent of the 60,000 legacy COBOL programs in the banking industry will still be in use by 2030. By providing a tool that can automate parts of the migration, OpenAI claims it can cut development time and costs by up to 40 percent, according to internal benchmarks.
However, experts caution that the technology is not a silver bullet. „Automated translation is helpful, but human oversight remains essential,” warned James Patel, a software architect at a leading investment bank. „Legacy systems often embed domain knowledge that a model may miss without contextual guidance.”
The Future of Legacy System Modernization
The question of adoption hinges on trust and regulatory compliance. Banks must ensure that any automated transformation meets stringent audit and security standards. OpenAI has stated that its models can be run on private servers, allowing institutions to keep code and data on premises. The company also offers a „code review” feature that flags potential security vulnerabilities in the generated code.
Some institutions are already piloting the technology. A mid‑size regional bank in the Midwest reportedly used the model to rewrite a COBOL batch job that processed mortgage payments. The pilot reported a 30 percent reduction in turnaround time for the migration, with no disruption to daily operations.
Frequently Asked Questions
If widely adopted, AI‑enhanced migration could reshape how banks handle legacy infrastructure. Faster, cheaper conversions may enable banks to adopt cloud-native architectures sooner, improving scalability and resilience. Moreover, freeing up legacy programmers to focus on higher‑value tasks could extend the useful life of their expertise.
Yet, the transition will not be without challenges. Regulatory bodies will need to scrutinize the automated processes, and banks must invest in training staff to work with AI‑generated code. The long‑term success of this approach will depend on balancing speed, accuracy, and compliance.
