Skip to content

Unicode perturbation families implemented as system-prompt soft barriers

Design · All designs
Evidence · 1 claim (1 for) · 1 study (1 causal), q2 · 0 of 1 report an effect size · 1 claim rests on one study

Description

The paper implements soft barriers as a system prompt instructing the model to inject specific Unicode characters into user-defined identifiers during generation, leaving keywords, operators, literals, and built-in names unchanged. Four families are evaluated: "Invisible chars" (zero-width characters inside identifiers), "Homoglyphs" (ASCII letters replaced by visually identical Unicode equivalents), "BIDI reordering" (directionality overrides altering internal byte order), and "Deletions" (backspace/delete controls erasing adjacent rendered characters). The mechanism requires no model retraining or environment changes.

Design Implications

Context

Requirements

  • Only a system prompt; no model retraining, decoder change, or execution-environment modification

Constraints

  • Any real deployment should be institutionally approved, disclosed at the policy level, and accessibility-tested; motivated users can bypass via clean-output requests, text normalizers, or retyping

Target Learners

  • programming students using AI coding assistants

Learning Goals

  • discouraging unexamined transfer of AI-generated code
  • promoting reading, editing, and reconstruction of generated code

Claims

Examples

-

Key Sources

  • Iyiola E. Olatunji, Alberick Euraste Djire, Jacques Klein, and Tegawendé F. Bissyandé. 2026. Do Not Copy/Paste: Soft Barriers for Copying in AI-Assisted Programming. Proceedings of the 41st IEEE/ACM International Conference on Automated Software Engineering (ASE '26). https://doi.org/10.1145/3832783.3834554
Page metadata
Field Value
type design
id unicode-perturbation-soft-barrier-families
title Unicode perturbation families implemented as system-prompt soft barriers
description The paper implements soft barriers as a system prompt instructing the model to inject specific Unicode characters into user-defined identifiers during generation, leaving keywords, operators, literals, and built-in na...
status review
generated by process:settle-candidates on 2026-10-11
trust unverified — no one has confirmed this page's content
sources 1 — iyiola-e-olatunji-alberick-euraste

Frontmatter is the page's machine-readable record — see the Schema & Guide.