When feeding sensitive information to AI, audit logs that track “who submitted what, and when” are essential. However, Claude’s Team plan does not include audit logs. Anthropic’s official help center explicitly states: “Audit logs are available for Enterprise organizations only.”

This post explores how to achieve audit log capabilities on the Team plan without upgrading to Enterprise.

What Enterprise Audit Logs Record

Before exploring workarounds, it helps to understand what the Enterprise plan’s audit logs actually capture.

Enterprise audit logs allow exporting up to 180 days of events as CSV. Organization Owners or Primary Owners can trigger an export from Admin Settings > Data and Privacy > Export logs, which sends a download link via email.

Each log entry contains the following fields:

Recorded events include SSO login/logout, user invitations/deletions, project creation/deletion/visibility changes, conversation creation/deletion, and file uploads, roughly 30 event types in total.

Notably, Enterprise audit logs are event logs that record “who did what, when.” Actual user-AI conversation content is not included. Only unique conversation identifiers are recorded. The Compliance API also explicitly excludes inference (conversation) activity from its logs.

To retrieve conversation inputs and outputs, you need the “data export” feature available to Primary Owners. However, this feature is also available on the Team plan. In other words, upgrading to Enterprise provides no additional capability for retrieving conversation content.

Feature Gap Between Team and Enterprise

Based on Claude’s pricing page, here is the feature gap around audit logging:

While data export and usage analytics are available on Team, the ability to track “who performed what action, when, from which IP and device” is Enterprise-only.

What Team Plan Can Natively Provide

The following data sources are available on Team and serve as building blocks for audit-like visibility.

Usage Analytics and Spend Report CSV

The Analytics dashboard provides per-user, per-model, per-day request counts, token usage, and estimated spend as CSV exports. This is useful for detecting anomalous usage patterns, though there is roughly a one-day delay and the data lacks operation-level granularity.

Organization Data Export

Primary Owners can export conversation data and user data in bulk from the Data and Privacy settings. However, this is not suitable for real-time monitoring, and deleted conversations are not included.

OpenTelemetry (Claude Code CLI / Cowork)

OpenTelemetry (OTel) monitoring is the most practical way to secure audit trails on the Team plan. It is available for both Claude Code CLI and Cowork (the desktop app’s agent feature, currently in Research Preview).

Cowork is currently in Research Preview and excluded from Enterprise features (audit logs, Compliance API, data exports), but OTel monitoring is available. Anthropic explicitly states OTel is “not a replacement for audit logs,” but on the Team plan, it provides the most granular audit trail available.

Collectable Data

The prompt.id attribute creates a traceability chain from user prompt to tool calls to API requests, enabling full tracking of “who did what, when, and with which tool.”

Admin Configuration: Claude Code CLI

Deploy managed-settings.json via MDM. Users cannot override these settings.

{
  "env": {
    "CLAUDE_CODE_ENABLE_TELEMETRY": "1",
    "OTEL_METRICS_EXPORTER": "otlp",
    "OTEL_LOGS_EXPORTER": "otlp",
    "OTEL_EXPORTER_OTLP_PROTOCOL": "grpc",
    "OTEL_EXPORTER_OTLP_ENDPOINT": "http://collector.example.com:4317",
    "OTEL_EXPORTER_OTLP_HEADERS": "Authorization=Bearer example-token"
  }
}

Admin Configuration: Cowork

In the desktop app, go to Organization settings > Cowork > Monitoring to configure OTLP endpoint, protocol, and headers. Once set by an admin, the configuration applies to all organization members.

Case Study: ZOZO

ZOZO published a case study on building a log collection system with Claude Code + OTel + BigQuery. They deploy managed-settings.json to all employee devices via Intune and route data through an OTel Collector on Cloud Run into BigQuery.

3-Layer Architecture for Audit Logging

No single tool covers everything. Combining three layers provides practical audit coverage on the Team plan.

flowchart LR
  U[Users] -->|SSO| IDP[IdP Logs]
  U -->|API usage| GW[API Gateway]
  GW --> CLAUDE[Claude API]
  U -->|Direct web usage| BM[Browser Monitoring]
  BM --> CLAUDE_WEB[Claude Web UI]
  U -->|Network| NW[CASB / Proxy]
  IDP --> SIEM[SIEM / Log Platform]
  GW --> SIEM
  BM --> SIEM
  NW --> SIEM

Layer 1: API Gateway

The most effective approach is to route Claude usage through an API gateway that intercepts all requests and responses. This captures full prompt text, full response text, token counts, cost, timestamps, and user metadata.

Here is a comparison of major tools:

Example LiteLLM Proxy configuration:

# litellm_config.yaml
model_list:
  - model_name: claude-sonnet-4-5
    litellm_params:
      model: anthropic/claude-sonnet-4-5-20250929
      api_key: os.environ/ANTHROPIC_API_KEY
general_settings:
  database_url: "postgresql://user:pass@host:5432/litellm"
  store_prompts_in_spend_logs: true  # Record all prompts and responses

For all of these tools, onboarding is as simple as changing ANTHROPIC_BASE_URL to the gateway endpoint in your existing application code.

Layer 2: Browser / Endpoint Monitoring

API gateways are effective for programmatic usage, but they cannot capture direct usage of claude.ai’s web UI. Browser extensions and endpoint monitoring tools fill this gap.

For small teams, PromptRail is the lowest-friction option. Organizations with MDM infrastructure in place will find Harmonic Security more practical.

Layer 3: Network Controls (CASB / DLP)

Organizations with existing CASB or SASE solutions can extend their infrastructure to monitor Claude usage at the network level.

• Netskope: DLP controls, activity-based policies, AI threat framework mapping
• Zscaler: Supports tenant restrictions via header injection
• Microsoft Purview DLP: Included in M365 E5. Endpoint DLP for monitoring sensitive data sharing with AI platforms

Network-level controls can determine “who accessed which service” but lack visibility into conversation content. They should be combined with API gateways and browser monitoring.

Tenant restrictions are also critical. Team plan does not natively support tenant restrictions, meaning users can switch to personal accounts and bypass organizational security policies. By having network proxies inject the anthropic-allowed-org-ids header, you can block all Claude access except through authorized organization IDs, ensuring all usage remains within the auditable environment.

Recommended Patterns by Organization Size

Pattern A: Small Teams (under 50 users)

Build baseline audit coverage with Team plan features and SSO logs.

Pattern B: Mid-size Teams (50-150 users)

Add API gateway and OpenTelemetry to Pattern A for visibility into “what was submitted.”

Pattern C: Large Organizations / Regulated Industries

Consider migrating to Enterprise. Anthropic introduced a self-serve Enterprise plan in February 2026, available from 20 seats with credit card billing. This provides native audit logs, the Compliance API, and custom data retention controls with significantly lower management overhead.

Log Management Best Practices

Collecting audit logs is only half the battle. Logs themselves become a concentration of sensitive data and require proper protection.

Communicating to IT Security Teams

Instead of Enterprise audit logs, the following compensating controls can provide equivalent coverage:

1. Route sensitive work through Claude Code CLI / Cowork where OTel provides audit trails
2. Prohibit sensitive data input via web chat and codify this in your AI usage policy
3. Export Claude Team usage analytics CSV for periodic per-user token consumption reports
4. Establish an AI Acceptable Use Policy defining data classification tiers and usage rules

The web chat gap in OTel coverage is addressed through policy controls. By routing sensitive work to Claude Code CLI and Cowork, you maximize the coverage of technical audit trails.

Conclusion

Building audit log capabilities on Claude Team requires a multi-layered approach centered on OTel monitoring.

Instead of Enterprise audit logs, OpenTelemetry monitoring enables Team plan users to secure audit trails for development work. OTel is available for both Claude Code CLI and Cowork, capturing prompt content, tool execution, cost, and user identity in real time.

To extend coverage further, combine with these three layers:

1. API gateway (LiteLLM, Helicone, Portkey, etc.) to log API-routed usage
2. Browser / endpoint monitoring (Harmonic Security, PromptRail, etc.) to capture direct web UI usage
3. Network controls (CASB / proxy) to enforce tenant restrictions and prevent personal account bypass

For cost efficiency, the open-source LiteLLM Proxy + Langfuse combination offers the lowest barrier to entry. For organizations with 20+ seats, migrating to the self-serve Enterprise plan is the most straightforward path.

The right choice depends on the granularity of logs your audits require, your team’s usage patterns (API-centric vs. web UI-centric), and how well the solution integrates with your existing security infrastructure.

That’s all for achieving audit log capabilities on Claude Team without Enterprise. From the gemba.

References

• Claude Code Monitoring Documentation
• Claude Code ROI Measurement Guide (Docker Compose setup)
• Claude Code Monitoring with Amazon Bedrock
• ZOZO Case Study: Claude Code Log Collection
• Claude Team Usage Analytics