Advanced Smart Contract Vulnerability Detection Using Large Language Models
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
With the rapid expansion of using smart contracts, protecting the security of these contracts has become crucial. Existing analysis tools for detecting vulnerabilities in smart contracts are unreliable as they often fall short in accuracy, primarily due to their low recall rates-a significant challenge in this field. In this work, we utilize the open-source SolidiFi benchmark dataset to detect vulnerabilities related to Integer overflow/underflow (IoU), reentrancy (RE), and timestamp dependency (TD). These contracts, verified and available on Etherscan, proved unsuitable for direct application of LLMs due to comments, functions, and variables that might reveal the nature of the vulnerabilities. To address this, we performed several preprocessing steps to prepare the dataset for further research. We utilize a large language model to identify vulnerable code, provide reasoning for the vulnerabilities, explain how an attacker might exploit them, and propose fixed code. We design our prompts using chain-of-thought and expert patterns. Finally, we evaluate the results using various metrics and expert reviewers to assess the correctness of the reasoning, potential security risks, and code fixes. Our experiments demonstrate that our approach outperforms existing tools and methods. Notably, our recall rates are significantly high-93.5%, 95.4%, and 93.8%-addressing the challenge of low recall in detecting IoU, RE, and TD vulnerabilities, respectively.
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.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.001 | 0.000 |
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 it