What happened
In a Nov 2, 2026 technical launch, Datadog introduced AI Guard, an inline runtime security product that discovers agents, evaluates every prompt/tool call/MCP interaction against session history and policy, blocks prompt injection, tool misuse and data exfiltration mid-loop, and masks 400+ types of secrets/PII in real time. It enforces via the existing Datadog tracer (one-click) or an AI gateway, and maps findings to OWASP Agentic Top 10 and MITRE ATLAS.
Why it matters
Datadog is bringing agentic runtime security into the observability stack many enterprises already run, offering inline intent-based interception (catching indirect injection that permission checks miss) without a new deploy. It signals observability vendors competing directly with standalone agent-guardrail tooling.
Applicability
Datadog customers shipping agents in production should trial AI Guard especially for agents consuming untrusted content; DevSecOps teams can enforce policies with the tracer already deployed. Evaluate now given the inline containment value.