Harmonic now supports OpenTelemetry

Harmonic now ingests OpenTelemetry data from a growing set of AI tools and analyzes it with the same small language models that power the rest of our platform. Security teams get prompt-level visibility, sensitive data detection, and Usage Intelligence for AI activity that happens outside the browser.

At launch, our OTel support covers Claude Cowork, Claude Code CLI, OpenAI Codex CLI, and OpenAI Codex Desktop. We expect that list to grow quickly as more vendors adopt the standard.

Why OpenTelemetry

OpenTelemetry is an open standard for software observability. A growing list of AI vendors, including Anthropic and OpenAI, emit OTel event data that captures what users do inside the application, including the prompts they submit. Any tool that supports OTel can be configured to send that data to a compatible endpoint. Harmonic operates one.

For some tools, OTel is the only path to prompt-level visibility. Claude Cowork is the clearest example. Anthropic's Compliance API does not cover Cowork activity, and neither do audit logs or data exports. OTel is the only mechanism Anthropic offers for seeing what users actually share with Cowork. With this release, Harmonic gives security teams full visibility into Cowork prompts, classified, identity-enriched, and integrated with the rest of their AI governance program.

What you actually get

A raw OTel feed pointed at a SIEM gives you a stream of prompt events. That's operational data, not insight. There's no classification, no sensitive data analysis, no connection to the rest of your AI governance program.

Harmonic processes the OTel stream the same way it processes browser activity. Prompts are grouped into tasks. Tasks roll up into use cases. Sensitive data exposures surface as alerts with full context. Identity enrichment ties every event back to the employee behind it.

Concretely, that means:

• Sensitive data alerts in Claude Code, Claude Cowork, and OpenAI Codex, tied to employee identity and full prompt context
• Usage Intelligence coverage that includes desktop and CLI activity, so your view of how the organization uses AI isn't missing the engineering team
• A single console and a single policy engine across everything, with no separate dashboard for "the desktop stuff"

‍

Where OTel fits in the platform

OTel is one of four data sources Harmonic uses to govern AI across the enterprise. Every event flows through the same detection pipeline and surfaces in the same portal, regardless of where it came from.

• Browser extension covers 1,000+ web-based AI tools
• MCP Gateway covers agentic workflows and tool calls
• Endpoint Agent covers 100+ desktop apps, CLIs, and AI-native IDEs, plus direct calls to inference providers
• OpenTelemetry adds prompt-level visibility for AI tools that emit it natively, including Claude Cowork, Claude Code, and OpenAI Codex

Why this matters

For the last two years, most AI governance programs have been built around the browser, because that's where AI lived. That assumption is breaking down. AI is now embedded in desktop applications, command-line tools, IDEs, and collaboration surfaces that never touch a web page. Employees move fluidly between them, often using the same AI provider in three or four different forms over the course of a day.

Development is one of the most visible examples. Source code, architecture decisions, customer data inside debugging sessions, and credentials pasted into a terminal all flow through tools like Claude Code and OpenAI Codex. But it isn't only engineering. Knowledge workers across functions are adopting desktop and CLI AI tools for the same reason developers did, because the experience is faster and more capable than the browser equivalent.

Governance has to follow the work. A program that only sees browser activity will miss a growing share of how the organization actually uses AI, and the share that's missed tends to involve the most sensitive material. OTel support is how we keep coverage complete as AI moves into the surfaces where the real work happens.

‍

Harmonic Security empowers safe AI adoption everywhere your employees need it. For further exploration, check out our securing Claude Cowork guide and our free, open-source Claude auditing tool.

‍