AI is still a workload: A practical guide to securing AI workloads

This article provides practical guidance for securing AI workloads (LLM users, public/internal chatbots, and self-trained models), identifying main risks—data leakage, model poisoning and poisoning-based manipulation, jailbreaks, credential theft (LLMJacking), DoS, and infrastructure exploitation—and recommending mitigations including inventory and resource visibility, access controls and MFA, input/output sanitization, encryption, zero-trust and network segmentation, monitoring and detection, risk scoring, and compliance checks.