AI migration of COBOL to Java preserves bugs, raising quality concerns
3 August 2026 · filed under a702bd373608
Bulletin, Specola Galactica Telescope Log — 3 August 2026
A report circulated on Hacker News, citing a paper posted to arXiv (2607.28271), describes an assessment of AI-driven migration of legacy COBOL programs to Java. According to the summary under discussion, the migration process preserved defects present in the original COBOL codebases rather than correcting them during translation.
The paper’s title, as listed, is “AI migrated legacy COBOL programs to Java, bugs included.” No further detail on methodology, sample size, or the specific systems tested was available in the material reviewed for this log.
The finding bears on a live question in software engineering: whether large-scale automated code translation tools, increasingly used to modernize decades-old enterprise systems written in COBOL, introduce new risks by faithfully reproducing latent bugs rather than flagging or fixing them. Institutions and firms relying on such tools to retire legacy mainframe systems have generally treated automated translation as a way to reduce both cost and error, not merely to change syntax.
The Specola notes this report as a data point in an ongoing discussion about the reliability of AI-assisted code migration, without independent verification of the arXiv paper’s claims beyond what was stated in the circulated summary. Readers seeking technical detail are directed to the original posting.
End of entry.