Pillar Security Unveils New AI Attack: BadEngram
News related to:Pillar Security · 1 min read
TEL AVIV, Israel, Oct. 13, 2026 /CourierPR/ -- Pillar Security, in collaboration with Fujitsu researchers, has uncovered a new supply-chain attack called BadEngram, which can implant a hidden, persistent backdoor in open-weight AI models by modifying only the memory component of the model. The discovery highlights a significant security risk in these models, which are increasingly favored by businesses due to their cost-effectiveness and transparency.
BadEngram works by altering only the memory-specific parameters of an AI model, leaving the rest of the model unchanged. This makes the backdoor difficult to detect through conventional model testing. The attack is particularly concerning because it can remain dormant until triggered by a short, arbitrary trigger.
The findings underscore the need for organizations to reassess the integrity and provenance of every learned component capable of influencing model behavior. As AI models evolve, the security boundary must expand to include the entire AI system, encompassing memory and software components. This is a critical step in ensuring the safety and reliability of AI systems, especially as businesses increasingly rely on open-weight models.
Ariel Fogel, an AI Security Researcher at the Office of the CTO at Pillar Security, emphasized the importance of validating the safety of deployed models.
Satoshi Imai, Head of Security Science Laboratory at Fujitsu Research, highlighted the broader implications of the research.
Pillar Security, a company that protects the AI lifecycle from development through deployment, recommends treating the complete model as a supply-chain artifact. Security controls should verify the integrity and provenance of memory tables, gates, projections, and other learned components, especially when models are converted, quantized, repackaged, or obtained through third-party repositories.
The discovery of BadEngram is a significant development in the field of AI security. Pillar Security, a Pioneer in the 2026 Gartner® Emerging Market Quadrant for AI Application Security and one of CRN’s 10 Hottest AI Security Companies, continues to lead the way in safeguarding AI systems against emerging threats.