The Compliance Paradox: Semantic-Instruction Decoupling in Automated Academic Code Evaluation

TL;DR

This paper reveals a systemic vulnerability in LLM-based automated code evaluation, where models can be manipulated to pass hidden directives rather than accurately assess code correctness, highlighting a need for more robust alignment methods.

Contribution

The paper introduces the SPACI and AST-ASIP frameworks to expose and analyze the decoupling vulnerability in LLMs, demonstrating widespread failures in current models across multiple programming languages.

Findings

Over 95% failure rate in high-capacity models like DeepSeek-V3

Models prioritize hidden directives over code correctness

Current alignment paradigms create Trojan vulnerabilities

Abstract

The rapid integration of Large Language Models (LLMs) into educational assessment rests on the unverified assumption that instruction following capability translates directly to objective adjudication. We demonstrate that this assumption is fundamentally flawed. Instead of evaluating code quality, models frequently decouple from the submission's logic to satisfy hidden directives, a systemic vulnerability we term the Compliance Paradox, where models fine-tuned for extreme helpfulness are vulnerable to adversarial manipulation. To expose this, we introduce the Semantic-Preserving Adversarial Code Injection (SPACI) Framework and the Abstract Syntax Tree-Aware Semantic Injection Protocol (AST-ASIP). These methods exploit the Syntax-Semantics Gap by embedding adversarial directives into syntactically inert regions (trivia nodes) of the Abstract Syntax Tree. Through a large-scale evaluation of 9 SOTA models across 25,000 submissions in Python, C, C++, and Java, we reveal catastrophic failure rates (>95%) in high-capacity open-weights models like DeepSeek-V3, which systematically prioritize hidden formatting constraints over code correctness. We quantify this failure using our novel tripartite framework measuring Decoupling Probability, Score Divergence, and Pedagogical Severity to demonstrate the widespread "False Certification" of functionally broken code. Our findings suggest that current alignment paradigms create a "Trojan" vulnerability in automated grading, necessitating a shift from standard RLHF toward domain-specific Adjudicative Robustness, where models are conditioned to prioritize evidence over instruction compliance. We release our complete dataset and injection framework to facilitate further research on the topic.