Notifications
Clear all
Topic starter
18/08/2026 2:34 pm
TL;DR: AI coding agents need telemetry, redaction, and audit trails because prompts, tool calls, and outputs create security, compliance, and cost questions that standard developer workflows do not answer, especially when agents can act directly on environment changes, according to Bindplane. The governance problem is not the model itself but the lack of durable evidence and least-privilege control over agent action.
NHIMG editorial — based on content published by Bindplane: monitoring Claude Code with OpenTelemetry for security, adoption, and spend
By the numbers:
- Only 5.7% of organisations have full visibility into their service accounts.
- 96% of organisations store secrets outside of secrets managers in vulnerable locations including code, config files, and CI/CD tools.
- 91.6% of secrets remain valid five days after the targeted organisation is notified, showing a critical gap in remediation procedures.
Questions worth separating out
Q: What fails when AI coding agents are not logged like privileged systems?
A: Teams lose the ability to reconstruct what the agent saw, what it changed, and whether a human approval was valid.
Q: Why do AI systems create identity and access risk beyond traditional AppSec?
A: Because AI systems often act through delegated access.
Q: What do security teams get wrong about telemetry for AI agents?
A: They often treat telemetry as a troubleshooting tool instead of a governance control.
Practitioner guidance
- Instrument agent runtimes with immutable approval logging Capture prompts, tool calls, responses, and approval events in a durable record that can be queried during incident response and audit.
- Apply pre-ingest redaction to sensitive telemetry fields Remove or mask personal data, secrets, and internal identifiers before telemetry leaves the controlled pipeline.
- Treat approved tool calls as privileged actions Bind each approval to a user, device, context, and destination so the organisation can distinguish normal assistance from unsafe execution.
What's in the full article
Bindplane's full article covers the operational detail this post intentionally leaves for the source:
- The exact OpenTelemetry collection pattern used to ingest Claude Code telemetry from developer endpoints
- The settings.json configuration and routing logic used to push telemetry to multiple destinations
- The redaction and filtering steps applied before data reaches storage and dashboards
- The practical split between operational metrics, full-fidelity audit traces, and developer experience analytics
👉 Read Bindplane's analysis of monitoring Claude Code with OpenTelemetry →
Claude Code telemetry and audit trails: are your controls ready?
Explore further
18/08/2026 3:20 pm
AI coding agents are becoming governed execution surfaces, not just productivity tools. Once a model can read prompts, call tools, and influence code paths, the security question shifts from model output quality to runtime accountability. That is an identity problem as much as an AI problem, because the agent's approved actions become delegated acts that require traceable authority. Practitioners should treat agent telemetry as governance evidence, not optional observability.
A question worth separating out:
Q: How should organisations govern tool approvals in AI coding workflows?
A: They should treat each approval as a privileged event with identity binding, context, and retention. That means recording who approved the action, what tool was used, where the action went, and whether the request matched policy before execution completed.
👉 Read our full editorial: Claude Code telemetry changes how teams govern AI coding agents
