Malice in Agentland: Down the Rabbit Hole of Backdoors in the AI Supply Chain
While finetuning AI agents on interaction data—such as web browsing or tool use—improves their capabilities, it also introduces critical security vulnerabilities within the agentic AI supply chain.
Authors
- Alexandre Drouin (ORCID: https://orcid.org/0000-0001-7718-0319)
- Nicolas Chapados (ORCID: https://orcid.org/0000-0003-0249-7607)
- Alexandre Lacoste
- Krishnamurthy Dvijotham (ORCID: https://orcid.org/0000-0002-1328-4677)
- Quentin Cappart (ORCID: https://orcid.org/0000-0002-8742-0774)
- Léo Boisvert
- Abhay Puri
- Chandra Kiran Reddy Evuru
Institutions
- Mila - Quebec Artificial Intelligence Institute (CA)
- ServiceNow (United States) (US)
- Minnow Environmental (Canada) (CA)
- Polytechnique Montréal (CA)
Publication Details
- Published
- 2026-05-22
- DOI
- https://doi.org/10.1145/3786335.3813166
- Primary Topic
- Blockchain Technology Applications and Security
- Type
- article
- Field-Weighted Citation Impact
- 0.00