Can admins see what employees prompt in Claude Code, Cursor, Codex or Copilot?

the short answer

Partly, and it varies by vendor. Claude Enterprise’s Compliance API returns Claude Code transcripts, and Claude Code OpenTelemetry can include prompts if you opt in. Cursor’s audit logs exclude prompts and code, and its conversation export covers only Cloud Agents and Grok Bot. Codex and Copilot expose prompts through compliance or telemetry features. Agent actions are easier: hook-level policy logs record each gated tool call on every machine.

Claude Code: Enterprise Compliance API transcripts; OTel prompts opt-in
Cursor: Audit logs exclude prompts; capture IDE prompts with a beforeSubmitPrompt hook
Codex: ChatGPT Compliance API (codex_log); [otel] log_user_prompt
Copilot: Managed telemetry.capture.prompts (CLI, app, VS Code Agent Host)
Actions, all supported agents: Hook-level policy and session logs

The Question Admins Keep Asking

On the Cursor forum an admin wrote: “I am managing a team using Cursor, and I want to control or monitor employees’ prompt/response data. I’ve checked both the dashboard and the API but couldn’t find this capability.” Another asked for an audit trail of “agent tool call proposals and user approval decisions”, noting the Enterprise audit log excludes agent actions. The answer differs by vendor.

What Each Vendor Exposes

Agent	Prompts	Agent actions
Claude Code	Compliance API transcripts (Enterprise); OTel `user_prompt` with `OTEL_LOG_USER_PROMPTS=1`	OTel `tool_decision`, `tool_result`; commands with `OTEL_LOG_TOOL_DETAILS=1`
Cursor	Not in audit logs; OTel `conversation_content` covers Cloud Agents and Grok Bot only and is opt-in; a `beforeSubmitPrompt` hook can capture IDE prompts	OTel tool-call metrics and API, hook and plugin logs (Enterprise); hooks; AI Code Tracking gives line attribution
OpenAI Codex	ChatGPT Compliance API records under `conversation_message` and `codex_log` (coverage depends on workspace setup); `[otel] log_user_prompt = true`	`codex.tool_decision`, `codex.tool_result` events
GitHub Copilot	Managed `telemetry.capture.prompts` with `telemetry.enabled` and `endpoint` (CLI, Copilot app, VS Code Agent Host); API usage-record streaming with `body` (public preview, EMU and data-residency enterprises)	Hooks; enterprise audit log records agentic events like PRs, not local session content

Should You Capture Prompts?

Prompts contain code, customer data and sometimes secrets. Capturing them creates a sensitive dataset to protect and retain.
For most security and compliance questions, actions matter more than prompts: what ran, what was blocked, on which machine.
Tell employees what you capture. Monitoring rules differ by country; involve legal and HR before turning on prompt capture.

Capture Actions Consistently Across Agents

A hook-level policy client records every gated tool call and decision in the same shape for Claude Code, Codex, Cursor, Copilot CLI and the other supported harnesses. Know what each Failproof option sends: transcripts go to Cloud by default (--no-transcripts keeps them local and still reports decisions); Cloud Jev sends each checked tool call and the recent prompt unless --no-transcripts is set; and failproofai otel enable captures text unless you add --no-content.

FAQ

Is it legal to monitor employee prompts?

It depends on jurisdiction and your employment agreements. Get legal review and disclose monitoring before enabling it.

Can I see prompts from personal accounts?

No vendor gives you that. Pin logins to your organization in managed settings.

What is the least invasive useful option?

Action and decision logs without transcript content.

Sources

Checked against the sources below on October 21, 2026. Harness settings, hook capabilities and cloud features change between versions.