CalvinBackup/www-project-ai-testing-guide
1
0
Fork 0
mirror of https://github.com/OWASP/www-project-ai-testing-guide.git synced 2026-05-31 19:41:40 +02:00
Code Issues Packages Projects Releases Wiki Activity

Update AITG-INF-05_Testing_for_Fine-tuning_Poisoning.md

Browse Source
This commit is contained in:
Matteo Meucci
2025-11-23 13:45:48 +01:00
committed by GitHub
parent 401014af9f
commit 39b8438ada
1 changed files with 3 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files
Document/content/tests/AITG-INF-05_Testing_for_Fine-tuning_Poisoning.md
+3 -9
View File
@@ -31,15 +31,9 @@ This test identifies vulnerabilities arising from poisoning during fine-tuning,
- **Regular Red Teaming and Auditing**: Periodically conduct red teaming exercises where a dedicated team actively tries to poison the fine-tuning process. This helps uncover vulnerabilities in the MLOps pipeline before they can be exploited by real attackers.
### Suggested Tools
- **Adversarial Robustness Toolbox (ART)**
- Provides extensive tools for crafting poisoning attacks and implementing defenses, including data sanitization and activation-based detection.
- Tool Link: [ART on GitHub](https://github.com/Trusted-AI/adversarial-robustness-toolbox)
- **BackdoorBench**
- An open-source toolkit for systematic evaluation of backdoor attacks and defenses, including fine-tuning poisoning detection.
- Tool Link: [BackdoorBench on GitHub](https://github.com/SCLBD/BackdoorBench)
- **Cleanlab**
- A data-centric AI package that can automatically find and fix label errors in datasets, which is a common technique used in poisoning attacks.
- Tool Link: [Cleanlab on GitHub](https://github.com/cleanlab/cleanlab)
- **Adversarial Robustness Toolbox (ART)**: Provides extensive tools for crafting poisoning attacks and implementing defenses, including data sanitization and activation-based detection - [ART on GitHub](https://github.com/Trusted-AI/adversarial-robustness-toolbox)
- **BackdoorBench**: An open-source toolkit for systematic evaluation of backdoor attacks and defenses, including fine-tuning poisoning detection - Tool Link: [BackdoorBench on GitHub](https://github.com/SCLBD/BackdoorBench)
- **Cleanlab**: A data-centric AI package that can automatically find and fix label errors in datasets, which is a common technique used in poisoning attacks - [Cleanlab on GitHub](https://github.com/cleanlab/cleanlab)
### References
- OWASP Top 10 for LLM Applications 2025. "LLM04: Data and Model Poisoning." OWASP, 2025. [Link](https://genai.owasp.org/llmrisk/llm042025-data-and-model-poisoning/)