The Pandemic Paradox of AI: Contagious Backdoors and Silent Compromise in the Foundation-Model Supply Chain
International Conference on AI Research · 18.11.2026 · 7 (1) · 294-303
DOI: 10.34190/icair.7.1.5449 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Conference on AI Research, online erschienen am 18.11.2026. Verfasst von Oludolamu Ademola Onimole und 9 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).
Abstract (Originalsprache)
This position paper develops a theoretical threat model, "The Pandemic Paradox," examining the catastrophic risks posed by Contagious Backdoors within AI ecosystems. Conceptualized as a "COVID-like" cognitive malware, the threat is characterized by high transmissibility, asymptomatic incubation, and the potential for systemic collapse across the digital economy. Unlike conventional backdoors rooted in static code manipulation, this form of infection operates at the model level, contaminating the probabilistic distributions and weight parameters acquired during training. It propagates silently through shared datasets, model distillation, transfer learning, and collaborative AI research pipelines, transforming AI infrastructures into vectors of recursive compromise. Equally consequential, and more easily overlooked, is that such compromise need not manifest as failure at all: the same mechanisms can be turned to covert exfiltration and espionage, silently betraying the sensitive data entrusted to a model while its outputs remain entirely plausible. Such compromise generates cascading economic, legal, and geopolitical consequences, including profound challenges in attribution, accountability, and liability. To confront this existential risk, the paper advances a conceptual framework for Defensive AI Systems (DAIS) that shifts AI security from reactive protocols to disrupting the propagation paths through which compromise spreads. The DAIS framework is organized around three interdependent pillars: Anticipatory Defense, Adaptive Resilience, and Assured Governance. Together, these pillars reposition AI security as a continuous and systemic architecture of protection rather than a reactive compliance exercise. Anticipatory Defense integrates AI-driven threat intelligence, adversarial red teaming, and predictive anomaly detection to identify latent vulnerabilities before exploitation. Adaptive Resilience employs hybrid detection pipelines, shadow-model deployment, automated containment, and continual learning mechanisms to preserve operational integrity under zero-day and self-propagating attacks. Assured Governance establishes verifiable accountability structures through mechanisms such as a comprehensive AI Bill of Materials (AIBOM), cryptographic verification protocols, independent oversight, and auditable compliance infrastructures. Collectively, the framework advances a model of digital hygiene and resilient governance designed to secure the foundation-model supply chain against the propagation of persistent and potentially inheritable cognitive compromise.
Zitieren
APA
Onimole, O. A., Pekarova, S., Hattoh, D. K., Attachie, A. A., Oppong, A., Dove, D., Osamor, J., Hetzel, D. D., Potter, L., & Palmer, X. (2026). The Pandemic Paradox of AI: Contagious Backdoors and Silent Compromise in the Foundation-Model Supply Chain. International Conference on AI Research, 7(1), 294-303. https://doi.org/10.34190/icair.7.1.5449
Vancouver
Onimole OA, Pekarova S, Hattoh DK, Attachie AA, Oppong A, Dove D, et al. The Pandemic Paradox of AI: Contagious Backdoors and Silent Compromise in the Foundation-Model Supply Chain. International Conference on AI Research. 2026;7(1):294-303. doi:10.34190/icair.7.1.5449
BibTeX
@article{Onimole2026,
author = {Onimole, Oludolamu Ademola and Pekarova, Sarka and Hattoh, Dodzi Koku and Attachie, Angela Aseye and Oppong, Abigail and Dove, Dominique and Osamor, Jude and Hetzel, Danny “Danhammer” and Potter, Lucas and Palmer, Xavier-Lewis},
title = {The Pandemic Paradox of AI: Contagious Backdoors and Silent Compromise in the Foundation-Model Supply Chain},
journal = {International Conference on AI Research},
year = {2026},
volume = {7},
number = {1},
pages = {294--303},
doi = {10.34190/icair.7.1.5449}
}
EUVIMED