legacy-code

1 article
sort: new top best
clear filter
0 4/10

LLMs like Claude Opus 4.6 are becoming effective at finding hidden bugs in code—including decades-old assembly—by reasoning about control flow rather than pattern matching, but simultaneously create 1.7x more bugs than humans and pose risks to unpatchable legacy systems that bad actors could exploit at scale.

Claude Opus 4.6 Mark Russinovich Microsoft Azure Anthropic Matthew Trifiro Adedeji Olowe Lendsqr SpotBugs CodeQL Snyk Code GPT-4.1 Mistral Large DeepSeek V3 Mozilla Firefox Black Duck Signal NCC Group Ghidra CodeRabbit Daniel Stenberg cURL
zdnet.com · CrankyBear · 1 day ago · details · hn