AI / ML Security — The Threat Landscape

ML systems have a fundamentally different attack surface than classic apps. The model is data-derived and probabilistic; the training corpus, the weights, and the inference API are all attackable; and the trust boundary between instructions and data collapses entirely in large language models. This opener maps the territory — lifecycle attack points, the four attack tiers, the supply-chain blind spots, and where MITRE ATLAS, OWASP ML Top 10, and OWASP LLM Top 10 fit.