skip to main content
US FlagAn official website of the United States government
dot gov icon
Official websites use .gov
A .gov website belongs to an official government organization in the United States.
https lock icon
Secure .gov websites use HTTPS
A lock ( lock ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites.

Explore Research Products in the PAR
It may take a few hours for recently added research products to appear in PAR search results.


Title: Evaluation of Large Language Models on Code Obfuscation (Student Abstract)
Obfuscation intends to decrease interpretability of code and identification of code behavior. Large Language Models(LLMs) have been proposed for code synthesis and code analysis. This paper attempts to understand how well LLMs can analyse code and identify code behavior. Specifically, this paper systematically evaluates several LLMs’ capabilities to detect obfuscated code and identify behavior across a variety of obfuscation techniques with varying levels of complexity. LLMs proved to be better at detecting obfuscations that changed identifiers, even to misleading ones, compared to obfuscations involving code insertions (unused variables, as well as variables that replace constants with expressions that evaluate to those constants). Hardest to detect were obfuscations that layered multiple simple transformations. For these, only 20-40% of the LLMs’ responses were correct. Adding misleading documentation was also successful in misleading LLMs. We provide all our code to replicate results at https://github.com/SwindleA/LLMCodeObfuscation. Overall, our results suggest a gap in LLMs’ ability to understand code.  more » « less
Award ID(s):
2150145
PAR ID:
10498498
Author(s) / Creator(s):
; ; ;
Publisher / Repository:
AAAI
Date Published:
Journal Name:
Proceedings of the AAAI Conference on Artificial Intelligence
Volume:
38
Issue:
21
ISSN:
2159-5399
Page Range / eLocation ID:
23664 to 23666
Format(s):
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. ; ; ; ; ; (, OpenReview.net)
    Code Large Language Models (Code LLMs) have excelled at tasks like code completion but often miss deeper semantics such as execution effects and dynamic states. This paper aims to bridge the gap between Code LLMs' reliance on static text data and the need for semantic understanding for complex tasks like debugging and program repair. We introduce a novel strategy, monologue reasoning, to train Code LLMs to reason comprehensive semantics, encompassing high-level functional descriptions, local execution effects of individual statements, and overall input/output behavior, thereby linking static code text with dynamic execution states. We begin by collecting PyX, a clean Python corpus of fully executable code samples with functional descriptions and test cases. We propose training Code LLMs not only to write code but also to understand code semantics by reasoning about key properties, constraints, and execution behaviors using natural language, mimicking human verbal debugging, i.e., rubber-duck debugging. This approach led to the development of SemCoder, a Code LLM with only 6.7B parameters, which shows competitive performance with GPT-3.5-turbo on code generation and execution reasoning tasks. SemCoder achieves 79.3% on HumanEval (GPT-3.5-turbo: 76.8%), 63.6% on CRUXEval-I (GPT-3.5-turbo: 50.3%), and 63.9% on CRUXEval-O (GPT-3.5-turbo: 59.0%). We also study the effectiveness of SemCoder's monologue-style execution reasoning compared to concrete scratchpad reasoning, showing that our approach integrates semantics from multiple dimensions more smoothly. Finally, we demonstrate the potential of applying learned semantics to improve Code LLMs' debugging and self-refining capabilities. Our data, code, and models are available at: https://github.com/ARiSE-Lab/SemCoder. 
    more » « less
  2. ; ; ; ; ; (, Proceedings of the 2019 ACM SIGSAC Conference on Computer and Communications Security)
    This paper presents MalMax, a novel system to detect server-side malware that routinely employ sophisticated polymorphic evasive runtime code generation techniques. When MalMax encounters an execution point that presents multiple possible execution paths (e.g., via predicates and/or dynamic code), it explores these paths through counterfactual execution of code sandboxed within an isolated execution environment. Furthermore, a unique feature of MalMax is its cooperative isolated execution model in which unresolved artifacts (e.g., variables, functions, and classes) within one execution context can be concretized using values from other execution contexts. Such cooperation dramatically amplifies the reach of counterfactual execution. As an example, for Wordpress, cooperation results in 63% additional code coverage. The combination of counterfactual execution and cooperative isolated execution enables MalMax to accurately and effectively identify malicious behavior. Using a large (1 terabyte) real-world dataset of PHP web applications collected from a commercial web hosting company, we performed an extensive evaluation of MalMax. We evaluated the effectiveness of MalMax by comparing its ability to detect malware against VirusTotal, a malware detector that aggregates many diverse scanners. Our evaluation results show that MalMax is highly effective in exposing malicious behavior in complicated polymorphic malware. MalMax was also able to identify 1,485 malware samples that are not detected by any existing state-of-the-art tool, even after 7 months in the wild. 
    more » « less
  3. ; ; ; ; (, Proceedings of the 35th Annual Computer Security Applications Conference)
    Malware written in dynamic languages such as PHP routinely employ anti-analysis techniques such as obfuscation schemes and evasive tricks to avoid detection. On top of that, attackers use automated malware creation tools to create numerous variants with little to no manual effort. This paper presents a system called Cubismo to solve this pressing problem. It processes potentially malicious files and decloaks their obfuscations, exposing the hidden malicious code into multiple files. The resulting files can be scanned by existing malware detection tools, leading to a much higher chance of detection. Cubismo achieves improved detection by exploring all executable statements of a suspect program counterfactually to see through complicated polymorphism, metamorphism and, obfuscation techniques and expose any malware. Our evaluation on a real-world data set collected from a commercial web hosting company shows that Cubismo is highly effective in dissecting sophisticated metamorphic malware with multiple layers of obfuscation. In particular, it enables VirusTotal to detect 53 out of 56 zero-day malware samples in the wild, which were previously undetectable. 
    more » « less
  4. ; ; ; (, COLM)
    As large language models (LLMs) expand the power of natural language processing to handle long inputs, rigorous and systematic analyses are necessary to understand their abilities and behavior. A salient application is summarization, due to its ubiquity and controversy (e.g., researchers have declared the death of summarization). In this paper, we use financial report summarization as a case study because financial reports are not only long but also use numbers and tables extensively. We propose a computational framework for characterizing multimodal long-form summarization and investigate the behavior of Claude 2.0/2.1, GPT-4/3.5, and Cohere. We find that GPT-3.5 and Cohere fail to perform this summarization task meaningfully. For Claude 2 and GPT-4, we analyze the extractiveness of the summary and identify a position bias in LLMs. This position bias disappears after shuffling the input for Claude, which suggests that Claude seems to recognize important information. We also conduct a comprehensive investigation on the use of numeric data in LLM-generated summaries and offer a taxonomy of numeric hallucination. We employ prompt engineering to improve GPT-4's use of numbers with limited success. Overall, our analyses highlight the strong capability of Claude 2 in handling long multimodal inputs compared to GPT-4. The generated summaries and evaluation code are available at https://github.com/ChicagoHAI/characterizing-multimodal-long-form-summarization. 
    more » « less
  5. ; ; ; ; ; ; (, OpenReview)
    Code Large Language Models (Code LLMs) are being increasingly employed in real-life applications, so evaluating them is critical. While the conventional accuracy evaluates the performance of Code LLMs on a set of individual tasks, their self-consistency across different tasks is overlooked. Intuitively, a trustworthy model should be self-consistent when generating natural language specifications for its own code and generating code for its own specifications. Failure to preserve self-consistency reveals a lack of understanding of the shared semantics underlying natural language and programming language, and therefore undermines the trustworthiness of a model. In this paper, we first formally define the self-consistency of Code LLMs and then design a framework, IdentityChain, which effectively and efficiently evaluates the self-consistency and conventional accuracy of a model at the same time. We study eleven Code LLMs and show that they fail to preserve self-consistency, which is indeed a distinct aspect from conventional accuracy. Furthermore, we show that IdentityChain can be used as a model debugging tool to expose weaknesses of Code LLMs by demonstrating three major weaknesses that we identify in current models using IdentityChain. Our code is available at https://github.com/marcusm117/IdentityChain. 
    more » « less
  • Citation Formats
  • MLA
  • APA
  • Chicago
  • BibTeX
  • Save / Share this Record
  • Send to Email