The complete runtime path of a data item through an AI system, including tool calls, retrievals, delegations, processors, and cross-border hops. It is the most useful evidence unit when static deployment records no longer match how an agent actually behaves.
Expanded Definition
Data trajectory describes the end-to-end runtime movement of a data item as it is processed by AI systems, especially when an agent invokes tools, retrieves context, delegates tasks, or passes information across services and jurisdictions. Unlike a static data flow diagram, it captures what actually happened during execution rather than what designers expected to happen.
For NHIMG, the term is most useful where AI systems interact with NHI, secrets, regulated records, or customer data, because the evidence trail must show not only where data was stored, but which runtime decision points touched it. That makes data trajectory a governance concept as much as an engineering one, and it fits naturally with NIST Cybersecurity Framework 2.0 thinking about visibility, protection, and recovery. Definitions vary across vendors, but no single standard governs this yet, so teams usually build the concept from logging, tracing, and policy enforcement evidence.
The most common misapplication is treating a deployment diagram as proof of actual data handling, which occurs when runtime agent behaviour, tool use, or cross-border transfers are not independently recorded.
Examples and Use Cases
Implementing data trajectory rigorously often introduces logging overhead and privacy review friction, requiring organisations to weigh traceability against operational complexity and data minimisation obligations.
- An AI customer service agent retrieves account history, calls a ticketing tool, and forwards a summary to a human resolver, creating a trajectory that must be traceable across each hop.
- A coding agent reads secrets from a vault, writes to a repository, and triggers CI jobs, making the runtime path essential for proving whether credentials were exposed or merely referenced.
- A cross-border analytics workflow routes data through a regional processor before returning outputs to the primary environment, so the trajectory becomes evidence for transfer review and residency checks.
- A retrieval-augmented generation pipeline pulls from internal documents, then caches snippets in an application layer, which means the trajectory must distinguish transient retrieval from durable retention.
- An autonomous agent delegates part of a task to another agent or service, and the trajectory helps show which entity exercised execution authority at each stage, a concern that aligns with NIST Cybersecurity Framework 2.0 and evidence-based control monitoring.
Why It Matters for Security Teams
Security teams need data trajectory because AI incidents often cannot be explained from static architecture alone. When an agent chains retrieval, tool use, delegation, and external calls, the real risk is not just where data was intended to go, but where it actually travelled and what it touched. That is critical for identity-bound records, NHI governance, and agentic AI oversight, where execution authority can outpace human review.
Data trajectory also supports investigation, access review, and policy enforcement by connecting runtime events to control obligations. It helps answer whether sensitive data left approved boundaries, whether a tool call was excessive, and whether a delegated action inherited inappropriate access. For teams mapping AI operations to governance expectations, the trajectory becomes a practical evidence layer alongside NIST Cybersecurity Framework 2.0 and, where identity assurance is involved, identity verification records.
Organisations typically encounter data trajectory as a forensic necessity only after an agent has already routed sensitive data through an unexpected path, at which point the concept becomes operationally unavoidable to address.
Standards & Framework Alignment
This section maps relevant standards and security frameworks to the operational risks and controls described in this guidance.
OWASP Agentic AI Top 10 and OWASP Non-Human Identity Top 10 address the attack and risk surface, while NIST CSF 2.0, NIST AI RMF and NIST SP 800-63 set the governance and control requirements practitioners need to meet.
| Framework | Control / Reference | Relevance |
|---|---|---|
| NIST CSF 2.0 | DE.AE | Data trajectories are runtime events that must be detected and analyzed for anomalies. |
| NIST AI RMF | AIRMF governance emphasizes mapping, measuring, and managing AI system risks across lifecycle activity. | |
| OWASP Agentic AI Top 10 | Agentic AI security guidance focuses on tool use, delegation, and unintended data exposure paths. | |
| OWASP Non-Human Identity Top 10 | NHI governance depends on knowing how machine identities and secrets are used during runtime. | |
| NIST SP 800-63 | IAL2 | Identity assurance becomes relevant when trajectories include identity-linked records or verification steps. |
Preserve evidence linking runtime data handling to identity assurance where records are identity-bound.
