Use this taxonomy as a buyer checklist. It defines the minimum adversarial coverage your reliability platform should provide before you trust it in production.
Input perturbations at inference time to force misclassification or missed detection. Primary applicable domains: All domains.
Malicious training data insertion, including backdoor and label-flip behavior corruption. Primary applicable domains: All, especially Medical, Fintech, Energy.
Repeated API probing to reconstruct model behavior and enable stronger follow-on attacks. Primary applicable domains: Fintech, Surveillance cloud deployments.
Reconstruction of sensitive training data from model outputs or confidence patterns. Primary applicable domains: Medical, Fintech, Surveillance.
Natural-language instruction hijacking that bypasses guardrails in LLM-integrated systems. Primary applicable domains: Medical NLP, Fintech assistants, LLM pipelines.
Patches, textures, light, sound, and objects used to fool sensors in real environments. Primary applicable domains: Robotics, Surveillance, Space GNC.
False sensor data feeds including LiDAR, GPS/GNSS, RF, and related channels. Primary applicable domains: Robotics, Space, UAV and autonomous systems.
Manipulated telemetry/SCADA/smart-meter or transit data to corrupt downstream AI behavior. Primary applicable domains: Energy, Space, IoT-connected systems.
Hidden trigger behaviors implanted during training that activate adversarial outputs later. Primary applicable domains: All, especially federated settings.
Adversarial inputs crafted on surrogate models that transfer to black-box targets. Primary applicable domains: All domains.
Robusto cannot focus on image-only perturbations; practical coverage requires tabular, time-series, graph, text, and multimodal support.
High-risk verticals need simulation for both digital-only and sensor/physical attack pathways.
Extraction and inversion risks make model endpoint monitoring and query-pattern analysis mandatory.
Taxonomy-backed testing evidence should map to regulatory and audit expectations by domain.