GitHub introduced a managed telemetry option that adds three identity attributes to OpenTelemetry streams for Copilot: the GitHub login (user.name), the local OS account (process.user.name), and the machine name (host.name). The feature is disabled by default and is separate from the existing pseudonymous identifier.
New Identity Capture Controls
The setting telemetry.capture.identity is the only knob that injects identity data into OTel payloads. Because it is off by default, administrators must explicitly enable it in the enterprise managed settings. Once enabled, the three attributes are attached to every Copilot‑related span, allowing downstream analysis to map activity back to a specific user and host without pulling in the actual prompt or response content.
Granular Content Capture Options
In addition to identity, GitHub retains fine‑grained controls for five content classes, each also disabled by default:
prompts– user, system, and developer prompts plus input context.responses– assistant replies and reasoning data.toolArguments– raw arguments, commands, schemas, and identifiers for tools, plugins, and hooks.toolOutput– results from tools and any execution‑error details.policyDetail– raw policy information such as sandbox filesystem paths.
The legacy captureContent flag still exists as a shorthand for the five classes, but it never turns on identity capture. Managed values always win over local SDK settings or environment variables, ensuring a consistent enterprise policy.
Operational and Security Considerations
Adding identity fields to telemetry streams introduces new data handling requirements. Teams must verify that the collection endpoint is approved, encrypted, and protected against unauthorized access, because host and OS user names are now part of the payload. Existing OTel pipelines may need schema updates to accept the new attributes, and alerting rules should be reviewed to avoid accidental exposure of identity data.
From a compliance perspective, the ability to attribute usage without storing prompt content can simplify audit trails while respecting privacy constraints. However, operators should still assess retention policies for the added identifiers, especially in regulated environments.
Related CloudNinjas coverage: AI engineering.
What This Means For Practitioners
Evaluate whether identity capture aligns with your organization’s monitoring goals. If you need per‑user usage metrics, enable telemetry.capture.identity and keep all content classes disabled. Update your OTel collector configuration to ingest the new fields and ensure the telemetry sink meets security requirements. Finally, monitor the managed‑settings hierarchy to confirm that local overrides cannot unintentionally re‑enable content capture.
