Skip to main navigation Skip to search Skip to main content

CoProtector: Protect Open-Source Code against Unauthorized Training Usage with Data Poisoning

  • Zhensu Sun
  • , Xiaoning Du
  • , Fu Song
  • , Mingze Ni
  • , Li Li*
  • *Corresponding author for this work
  • Monash University
  • ShanghaiTech University
  • University of Technology Sydney

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Github Copilot, trained on billions of lines of public code, has recently become the buzzword in the computer science research and practice community. Although it is designed to help developers implement safe and effective code with powerful intelligence, practitioners and researchers raise concerns about its ethical and security problems, e.g., should the copyleft licensed code be freely leveraged or insecure code be considered for training in the first place? These problems pose a significant impact on Copilot and other similar products that aim to learn knowledge from large-scale open-source code through deep learning models, which are inevitably on the rise with the fast development of artificial intelligence. To mitigate such impacts, we argue that there is a need to invent effective mechanisms for protecting open-source code from being exploited by deep learning models. Here, we design and implement a prototype, CoProtector, which utilizes data poisoning techniques to arm source code repositories for defending against such exploits. Our large-scale experiments empirically show that CoProtector is effective in achieving its purpose, significantly reducing the performance of Copilot-like deep learning models while being able to stably reveal the secretly embedded watermark backdoors.

Original languageEnglish
Title of host publicationWWW 2022 - Proceedings of the ACM Web Conference 2022
PublisherAssociation for Computing Machinery, Inc
Pages652-660
Number of pages9
ISBN (Electronic)9781450390965
DOIs
StatePublished - 25 Apr 2022
Externally publishedYes
Event31st ACM Web Conference, WWW 2022 - Virtual, Lyon, France
Duration: 25 Apr 202229 Apr 2022

Publication series

NameWWW 2022 - Proceedings of the ACM Web Conference 2022

Conference

Conference31st ACM Web Conference, WWW 2022
Country/TerritoryFrance
CityVirtual, Lyon
Period25/04/2229/04/22

Keywords

  • data poisoning
  • dataset protection
  • deep learning
  • open-source code

Fingerprint

Dive into the research topics of 'CoProtector: Protect Open-Source Code against Unauthorized Training Usage with Data Poisoning'. Together they form a unique fingerprint.

Cite this