Least privilege
Issue short-lived, task-specific access instead of inheriting a user's full authority.
Enterprise Governance & Control
Strict sandboxing and comprehensive logging form the foundation of our architecture, minimizing deployment risks while elevating productivity.
Feature 01
Security & Infrastructure Isolation
Features strict, network-restricted, air-gapped deployment architecture and staging simulations.
STAGING_SIMULATOR: SECURED
PORT 3000 CONFINED
SECURITY_REHEARSAL_SUITE
TESTING ACTIVE
DRY_RUN REHEARSAL SUCCESSFUL
100% PASS
IMMUTABLE_AUDIT_LOG
LEDGER_SECURE
[02:31:02]
SYS_INIT: sandbox
booted
✓ signed
[02:31:05]
WRITE_OUT:
blocked (WAN)
✗ alert
[02:31:12]
HASH:
f99e3a32f012de9c
verified
CRYPTOGRAPHIC PROOF
READ ONLY
Air-Gapped Sandbox Environment with Network Control
Air-gapped deployment architecture with strict, real-time outbound network controls.
Automated Pre-Deployment Security Testing
Local dry-run rehearsal environment to deeply validate workflows before pushing them live.
Comprehensive Read-Only Audit Logs
Full, immutable audit trailing capturing all agentic operations and reads.
Point 02
Governance, Permissions & Human Oversight
Features focused on access control, strict corporate guardrails, and human-in-the-loop validation.
Claw Bot Channel
ENCRYPTED
CB
Here's your morning briefing:
- 3 tasks awaiting human approval
- 2 compliance flags identified
Do you want to schedule this tasks?
Schedule Task Options
Step-by-step log
Yes, set up due day
5 tasks reviewed
No, it is a single task
5 tasks reviewed
Settings
Security, governance, and system configuration
Staff Accounts
Team Admin
Security & Compliance
General
| Name / Email | Role | Department | Status |
|---|---|---|---|
|
SC
Sarah Chen
sarah.chen@openclaw.com
|
Admin | Operations | Active |
|
MT
Michael Torres
m.torres@openclaw.com
|
Staff | Sales | Active |
|
EW
Emily Wright
e.wright@openclaw.com
|
Custom Admin | Finance | Active |
|
JP
James Park
james.p@openclaw.com
|
Staff | Marketing | Suspended |
Skills & Permissions
Configure granular access controls for AI agents
Skills (5)
Plugins (5)
Templates
4 out of 5 skills are whitelisted with read/write
permissions configured
File Reader
READ ONLY
Whitelisted
Permissions
Path Restrictions
/company/public/**
×
/reports/readonly/**
×
+ Add Path
Database Writer
READ & WRITE
Whitelisted
Human-in-the-Loop Approval Mechanism
Immutable approval checkpoints preceding all external data writing and workflow execution tasks.
Granular Role-Based Access Control
Role-Based Access Control (RBAC) tightly locks down access permissions to specific user blocks.
Pre-Approved Allowlist of Available Skills
Secure, enterprise-grade capability skills pre-approved and locked down by super-admins.
Point 03
Resource Management & Centralized Operations
Features focused on monitoring costs, configuring AI behavior, and orchestrating tasks from a single interface.
Task Control Board
User:
Woody (RM)
W
4 tasks pending human approval
To Do
2
Investor Report
Report Agent
0/4 done
CRM Enrichment
CRM Agent
In Progress
1
ESG Compliance
Compliance
1 approval needed
Needs Approval
2
Q4 Portfolio
Report Agent
Pending...
Lead Qualification
Lead Agent
Completed
1
Daily Briefing
✓ 4/4 Tasks
Finished
Claw Bot Channel
15:57 PM
CB
Here's your morning briefing:
- 3 tasks awaiting approval
- 2 compliance flags on ESG
Do you want to schedule this tasks?
Yes, set up due day
No, single task
Quota Monitor
Live credit consumption
Credits Remaining
8,420
≈ $421 remaining
Daily Limit
10,000
Skills & Permissions
Configure granular access controls for AI agents
Skills (5)
Plugins (5)
Templates
4 out of 5 skills are whitelisted with read/write
permissions configured
File Reader
READ ONLY
Whitelisted
Permissions
Path Restrictions
/company/public/**
×
/reports/readonly/**
×
+ Add Path
Database Writer
READ & WRITE
Whitelisted
Unified Central Management Interface
Command-center UI to configure, orchestrate, and manage all background tasks and connections.
Token & Usage Management
Precise consumption monitoring and budgetary limits enforced on LLM compute pools.
Skill Configuration & Optimization Interface
Specialized administrator panel to rapidly tune, configure, and refine custom AI behaviors.
Security is an operating model
Control the agent's authority—not only its answers
Enterprise agents read records, call tools and can change business state. That expands the risk surface beyond content accuracy. Security therefore has to govern identity, data, actions, runtime isolation, third-party content and the people responsible for every consequential workflow.
A practical control stack for agentic work
Data boundaries
Classify what the agent may retrieve, retain, transform and disclose for each workflow.
Isolated execution
Run code, files and browser activity in controlled environments with network and resource limits.
Prompt-injection defence
Treat web pages, documents and messages as untrusted data—not instructions that can expand authority.
Approval policy
Require human review according to action risk, data sensitivity, value and reversibility.
Detection and response
Monitor anomalous behaviour, preserve traces, suspend agents and investigate with reproducible evidence.
The secure execution loop
- 01
Authenticate
Resolve the user, agent and service identities.
- 02
Authorise
Calculate permitted data, tools and actions for this task.
- 03
Isolate
Create a controlled workspace with explicit network and runtime limits.
- 04
Inspect
Screen external content and proposed actions for policy violations.
- 05
Approve
Escalate sensitive or irreversible steps to an accountable reviewer.
- 06
Observe
Record behaviour, outcome and anomalies for audit and response.
Risk-to-control map
| Risk | Enterprise control | Evidence for assurance |
|---|---|---|
| Excessive access | Scoped credentials and policy enforcement | Access decision and resource log |
| Prompt injection | Untrusted-content isolation and action review | Source trace and blocked instruction record |
| Wrong or incomplete result | Outcome tests and human exception queue | Evaluation score and reviewer decision |
| Unapproved external action | Approval gate and allow-listed destinations | Approval identity, time and payload |
| Runaway cost or activity | Time, token, tool and transaction budgets | Budget alerts and termination reason |
Questions for the CISO and risk committee
- 01
Can one compromised document cause the agent to acquire more authority?
- 02
Can administrators see every connected system, credential, skill and scheduled run?
- 03
Can a run be stopped centrally and reconstructed afterwards?
- 04
Are high-risk actions blocked by policy, not merely discouraged in a prompt?
- 05
Does the deployment model meet residency, retention and key-management requirements?
A secure enterprise agent is not an unrestricted digital employee. It is a constrained operator whose authority, environment and evidence are continuously governed.