Reproduce “Vibe Code Encryption: Adversarial Semantic Misdirection of Codebases Against AI Agent Reconnaissance” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Shubhankar Tripathy, Manu Prabakaran, Luis Sanchez, Viren Velacheri
Vibe Code Encryption (VCE) changes source-code names, structure, and documentation to steer unauthorized agents toward a decoy interpretation, while signed context helps authorized agents recover the intended meaning. In a paired test of 55 functions, Claude Sonnet 4.6’s truth-versus-decoy similarity margin fell from 0.317 to 0.232, but accuracy remained 100%, indicating margin erosion rather than frontier-model deception.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.