Are Classical Clone Detectors Good Enough For the AI Era?
Bibliographic record
Abstract
The increasing adoption of AI-generated code has reshaped modern software development, introducing syntactic and semantic variations in cloned code. Unlike traditional human-written clones, AI-generated clones exhibit systematic syntactic patterns and semantic differences learned from large-scale training data. This shift presents new challenges for classical code clone detection (CCD) tools, which have historically been validated primarily on human-authored codebases and optimized to detect syntactic (Type 1-3) and limited semantic clones. Given that AI-generated code can produce both syntactic and complex semantic clones, it is essential to evaluate the effectiveness of classical CCD tools within this new paradigm. In this paper, we systematically evaluate nine widely used CCD tools using GPTCloneBench, a benchmark containing GPT-3-generated clones. To contextualize and validate our results, we further test these detectors on established human-authored benchmarks, BigCloneBench and SemanticCloneBench, to measure differences in performance between traditional and AI-generated clones. Our analysis demonstrates that classical CCD tools, particularly those enhanced by effective normalization techniques, retain considerable effectiveness against AI-generated clones, while some exhibit notable performance variation compared to traditional benchmarks. This paper contributes by (1) evaluating classical CCD tools against AI-generated clones, providing critical insights into their current strengths and limitations; (2) highlighting the role of normalization techniques in improving detection accuracy; and (3) delivering detailed scalability and execution-time analyses to support practical CCD tool selection.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.009 | 0.076 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.003 | 0.003 |
| Science and technology studies | 0.001 | 0.002 |
| Scholarly communication | 0.004 | 0.006 |
| Open science | 0.004 | 0.002 |
| Research integrity | 0.002 | 0.002 |
| Insufficient payload (model declined to judge) | 0.002 | 0.002 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".