Anthropic says Claude can understand COBOL, IBM stock suffers worst fall in 25 years

IBM shares saw their sharpest fall in over 25 years after AI startup Anthropic said its Claude Code tool can understand and modernise COBOL, a language deeply tied to IBM's mainframe business. Here is everything that you need to know.

Shares of IBM suffered their steepest single-day fall in more than 25 years on Monday after fresh concerns surfaced around how AI could disrupt one of the company’s most entrenched businesses. The trigger came from AI startup Anthropic, which said its Claude Code tool is capable of understanding and modernising COBOL, a decades-old programming language that underpins many systems running on IBM’s mainframes.

Following this, IBM stock closed the session down 13.2 per cent at $223.35, which is its biggest daily drop since October 18, 2000. According to Reuters, the sell-off has dragged the stock down roughly 25 per cent so far this year, as investors reassess how quickly AI tools could alter the economics of enterprise software and IT services.

The market reaction followed a blog post from Anthropic claiming how Claude Code can automate large parts of COBOL modernisation, an area that has historically required long, consultant-heavy projects and generated steady revenue for IBM. The company argued that AI removes one of the biggest bottlenecks in updating legacy systems, which is simply understanding how they work.

“Hundreds of billions of lines of COBOL run in production every day, powering critical systems in finance, airlines, and government. Despite that, the number of people who understand it shrinks every year,” Anthropic wrote. “AI excels at streamlining the tasks that once made COBOL modernisation cost-prohibitive.”

COBOL, short for Common Business-Oriented Language, was created in the late 1950s and remains deeply embedded in global banking, insurance and government infrastructure. IBM has spent decades selling and supporting mainframe systems optimised for large-scale transaction processing, where COBOL continues to play a central role. Anthropic estimates that about 95 per cent of ATM transactions in the US still rely on the language, highlighting both its scale and its age.

Anthropic said Claude Code can analyse massive COBOL codebases by tracing dependencies across thousands of lines of code, generating documentation for workflows that are no longer clearly understood, and flagging risks that would otherwise take months of manual effort to uncover. “Legacy code modernisation stalled for years because understanding legacy code costs more than rewriting it. AI flips that equation,” the company said.