Editorial illustration for OpenAI reports promising results migrating COBOL code to modern languages
OpenAI Migrates Legacy COBOL Code to Modern Languages
The IRS runs on it. So do most banks, every major airline, and the systems that keep Social Security checks going out on time. COBOL, a programming language built in 1959 for business computing, has somehow outlived the companies that trained people to write it, the manuals that explained it, and the engineers who understood it.
IBM estimates COBOL still underpins more than 40 percent of online banking systems, 80 percent of in-person credit card transactions, and 95 percent of ATM transactions worldwide. Governments depend on it just as heavily.
Nobody wants to touch it. Rewriting decades-old financial or government code risks breaking something nobody fully understands anymore, and the programmers who built these systems are mostly retired or dead. That standoff has left agencies and banks stuck paying a steep price just to keep antique infrastructure limping along.
On October 19th, OpenAI said its newest AI models show "substantial improvements" at translating old code like COBOL into modern programming languages, a capability the company says could help organizations "maintaining and modernizing older systems." Whether AI can actually untangle seven decades of institutional code debt is the question now on the table.
Experts who spoke to The Verge agreed that AI tools have made strides in tackling legacy code in recent years. But they cautioned that advances trumpeted by OpenAI this week are a very long way from making it possible to safely replace critical legacy systems built around it.
Why this matters
COBOL isn't some museum relic. It still runs IRS filings, bank ledgers, and airline reservation systems that nobody wants to touch because one wrong move breaks something nobody remembers how to fix. OpenAI's claim that its models can migrate that code to Python or Rust while preserving behavior is the part worth watching closely, not the press release framing.
Preserving behavior is the entire ballgame here. A model that writes plausible-looking Kotlin from 40-year-old COBOL is useless, dangerous even, if it silently changes how interest gets calculated or how a flight booking gets processed. For developers and founders eyeing this space, the real test isn't benchmark scores on translation accuracy.
It's whether banks and government IT departments, the most risk-averse buyers on the planet, trust these tools enough to run them against production systems. OpenAI says the results are promising. That's a company talking about its own product.
We'd want to see independent validation on an actual legacy system before calling this solved, and so far that's not what's been offered.
Related Reading
Browse all OpenAI news →
Policy & Regulation Akamai Proposes SAFE Cybersecurity Guidelines for AI Threats
Policy & Regulation After ditching AI fitness apps and a Fitbit, I return to Peloton classes
Policy & Regulation Sony Music, Warner Chappell Sue Anthropic Over "Blatant Theft
Related Topic OpenAI researcher quits, citing distrust over ad‑driven engagement metrics
LLMs & Generative AI
Related Topic OpenAI launches GPT-Image 1.5 with precise editing for enterprise visuals
LLMs & Generative AI
From Archives Three fired OpenAI safety researchers say they were never told what they leaked
From Archives AI Makes Historic PHP Bugs More Exploitable, Turning Them Into Current Crisis
From Archives Eric Schmidt compares OpenAI and Anthropic to colonial trading companies
Business & Startups
From Archives OpenAI Discounts Poach Anthropic Clients; Rippling CEO Cites 90% Coding Share Shift
Business & Startups
Common Questions Answered
What percentage of critical financial systems still rely on COBOL according to IBM estimates?
IBM estimates that COBOL underpins more than 40 percent of online banking systems, 80 percent of in-person credit card transactions, and 95 percent of ATM transactions. This demonstrates that COBOL, despite being built in 1959, remains deeply embedded in essential financial infrastructure that processes trillions of dollars annually.
What is OpenAI's main claim about migrating legacy COBOL code to modern languages?
OpenAI claims that its AI models can migrate COBOL code to modern programming languages like Python or Rust while preserving the original behavior of the code. However, experts caution that these advances are still a very long way from making it safe to replace critical legacy systems that depend on COBOL.
Why is preserving behavior considered critical when migrating COBOL code to new languages?
Preserving behavior is essential because COBOL systems run mission-critical infrastructure like IRS filings, bank ledgers, and airline reservation systems where even a single error could cause major disruptions. Since many engineers no longer understand how the original COBOL code works, any behavioral changes during migration could break systems that nobody remembers how to fix.
What challenges do experts identify with using AI tools for COBOL code migration?
While experts acknowledge that AI tools have made recent strides in tackling legacy code, they caution that the advances claimed by OpenAI are preliminary and far from proven for real-world critical systems. The gap between demonstrating code migration capability and safely replacing systems that process billions of transactions remains substantial.
Further Reading
- Advancing coding in low-resource programming languages - OpenAI
- Can AI help solve the COBOL code crisis? - The Verge
- OpenAI looks to help banks modernise legacy software their operations still depend on - The Economic Times
- OpenAI's COBOL migration score has made a striking leap - The New Stack
