SALSA: Single-Pass Detection of Machine-Generated Code for SemEval-2026

A new paper on arXiv addresses the challenge of detecting machine-generated code, a growing concern as LLMs transform code generation. The work targets SemEval-2026 Task 13 Subtask A, which frames detection as a binary classification problem with emphasis on out-of-distribution generalization across unseen programming languages and application domains.

The proposed SALSA (Single-pass Autoregressive LLM Structured Classification) formulation enables code source detection in a single forward pass without additional training or complex post-processing, while maintaining strong cross-domain generalization.

Published on arXiv cs.CL on June 25, 2026, this research addresses practical needs in academic integrity, software supply chain security, and code provenance verification.