Shiva Raju

When the Observer Acts: Threat Modeling Agentic AI in Observability Pipelines

An explicit threat model is constructed that proposes twelve design-level controls organized into four families — provenance, corpus integrity, least authority, and accountability — and identifies the residual risks that no current control fully addresses.