Code reuse speeds up development but often goes unmanaged, leaving security holes in software. Pluvio, a new tool described on arXiv, aims to spot such reused code directly in assembly language, even when the code comes from architectures or libraries the model has never seen before.
The approach combines transfer learning with a conditional variational information bottleneck. This lets Pluvio match assembly clones by their underlying behavior rather than surface syntax, making it useful for auditing binaries where source code is unavailable or where the original build environment is unknown.
Because it works across out-of-domain settings, Pluvio could help security researchers trace vulnerable code snippets as they propagate through different software projects, even when those projects use different compilers, processors, or third-party libraries. The paper is a single source, so no conflicting findings are reported.