To Spot Attacks Through AI Models, Companies Need Visibility

Organizations rapidly adopting AI/ML face growing exposure from malicious or tampered models in public repositories and insecure model formats that enable code execution, backdoors, and biased outputs. Drawing on research and scans from JFrog, Protect AI, and HiddenLayer, the report highlights unsafe models, limited visibility into model inventories, and the need for MLSecOps: securing and tracking training data, generating AI bills of materials, and scanning and monitoring models and pipelines with dedicated tools to build trust and reduce attack surface.