Legacy Code That IRS and Banks Dare Not Touch: OpenAI Steps In With AI
Numerous core systems in global government, finance, and aviation sectors still run on decades-old legacy code like COBOL, which carries high migration risks and a shortage of maintenance personnel, making organizations generally reluctant to modify it easily. OpenAI’s recently announced large language model capabilities show substantial progress in migrating legacy code, aiming to tackle this tough problem of legacy systems.
The IRS, banks around the world, airlines, and various core government infrastructure systems share an unspoken problem. Their underlying systems rely on decades-old "antique" code, and no one dares to modify it casually for fear of triggering a chain reaction that could cause the entire business to collapse.
OpenAI thinks it has found a solution. On October 19, the company announced progress in the capabilities of its latest model, stating that it has made substantial improvements in migrating code written in outdated programming languages, able to convert legacy code from old software into modern programming languages, helping organizations in need complete the maintenance and modernization of old systems.
The language OpenAI focused on in this announcement is COBOL. Born in 1959, designed specifically for commercial data processing, it has long since ceased to be a mainstream programming language. Because core systems require extremely high stability, the cost and risk of full migration are unbearable, so many organizations have simply continued to use it to this day. The core system used by the IRS to process tax forms, bank fund settlement links, and airline booking systems all contain massive amounts of COBOL code.
The real-world dilemma has existed for many years. Programmers who mastered COBOL back then have retired one after another, and regular universities no longer offer relevant courses, so fewer and fewer people can maintain existing code. In the past, some organizations tried manual migration, with project cycles often taking several years, costs amounting to hundreds of millions of dollars, and compatibility issues popping up from time to time. Most organizations simply maintained the status quo, muddling through day by day.
It’s not new that large language models support writing code in mainstream programming languages, but being able to handle low-resource old languages like COBOL is a real industry problem. After all, these systems carry real-money transactions and citizen data, so no one dares to easily let AI touch core links.
At present, OpenAI has only announced progress in its capabilities, with no public cases of implementation partnerships. Whether these old systems that have run for more than half a century can be smoothly replaced depends on the accuracy of actual migrations and long-term stability.
发布时间: 2026-10-26 21:15