Loading...
Skip to Content

Non-Human Identity Governance

Non-Human Identity Governance

In the AI Agent era, "a legitimate account login" no longer means "a real human is operating." WuThreat ITDR Gateway 3.0 — without modifying business systems or installing any client — governs all invisible, uncontrolled, and untraceable non-human access through a 5-layer fingerprint + accountability tracing + app authorization matrix.

Full NHI Asset Inventory

Automatically discover and inventory all API keys, Service Accounts, AI Agent Tokens, and machine certificates enterprise-wide — eliminating non-human identity blind spots and establishing a single trusted asset ledger.

Automated Least-Privilege Control

Analyzes actual access behavior to automatically identify over-privileged accounts and intelligently recommend least-privilege policies — reducing lateral movement and privilege escalation risk.

Real-Time Anomaly Detection

Monitors non-human identity access behavior in real time — precisely identifying key leakage, abnormal call frequency, unauthorized time-window access, and permission violations with millisecond alert response.

5 Categories of Non-Human Identity

Full Coverage Across All NHI Types

From coding AI assistants embedded in developer IDEs, to enterprise RPA robots, to AI tools weaponized by external attackers — WuThreat classifies all non-human access into 5 categories, each with targeted identification and governance strategies.

Category 01 · Code AI

Code Generation AI

Cursor IDE · GitHub Copilot · Claude Code · Codeium

AI assistants embedded in dev environments that access code repositories, test APIs, or even production databases using developer credentials. Traditional IAM sees only "developer normal login."

Greylist · App Authorization Matrix
Category 02 · General Agent

General AI Agent

OpenClaw · Devin · Kimi · Custom Agents

Autonomous agents that hijack employee OAuth credentials for cross-system operations. Core risk: OAuth scope violations and autonomous decisions without human review.

Greylist · MFA Re-verification
Category 03 · Browser Agent

Browser Hijacking AI

OpenAI Operator · Browser Use · AutoGPT Browser Mode

Directly takes over real browser sessions, bypassing completed MFA. Traditional bot detection (honeypots) is nearly ineffective against these.

Strict Review · Behavioral Baseline
Category 04 · Enterprise RPA

Enterprise RPA

UiPath · Alibaba Cloud RPA · Custom Python Bots

Legitimate enterprise automation robots, often mistakenly flagged as attack sources. Require whitelist management and credential lifecycle auditing — not blocking.

Whitelist · Credential Audit
Category 05 · External Attack

External Attack Tools

Hermes Agent · LangChain Scrapers · Browser Use Credential Stuffing Scripts

Open-source AI tools repurposed as automated attack weapons. WAF sees only legitimate HTTPS traffic; credentials are real (obtained via stuffing). Completely invisible to WAF.

Blacklist · IP Range Block

Non-Human Identity Capabilities

NHI Governance Capabilities

From API keys to AI Agents — full lifecycle management for all non-human identity types enterprise-wide. Automated inventory, privilege governance, anomaly detection, and compliance auditing in one unified solution.

API Key Full Lifecycle Management

Automatically discovers API keys scattered across systems, centrally controls validity and permission scope, detects key leakage and abnormal calls in real time, and supports scheduled auto-rotation.

Service Account Privilege Governance

Full inventory of service account distribution and permission configuration — precisely identifies over-privileged accounts and zombie accounts, auto-recommends least-privilege policies with one-click decommissioning.

AI Agent Identity Access Control

Assigns independent least-privilege identities to each AI Agent, preventing injected malicious instructions from using Agent credentials to bypass business authorization or perform lateral penetration.

Zombie Credential Auto-Cleanup

Credentials unused for 90+ days are automatically flagged with warnings; expiring certificates trigger smart reminders — preventing stale credentials from becoming attackers' hidden entry points.

NHI Asset Overview

Monitoring
API Keys
Active: 893 · High-risk: 47 · Expired: 354
47 High-risk Total: 1,247
Service Accounts
Over-privileged: 23 · Zombie: 54 · Normal: 309
23 Over-privileged Total: 386
AI Agent Tokens
New (7d): 18 · High-privilege: 8
Governed Total: 142
Machine Certs / TLS
Expiring soon: 12 · Expired: 3 · Normal: 74
12 Expiring Total: 89

5-Layer Fingerprint Stack

Multi-Layer Identity Fingerprinting

A single fingerprint layer is easily forged; five stacked layers are not. L1 to L5 assess network, application, device, behavior, and identity dimensions jointly — even if an attacker bypasses the first three layers, behavioral layer L4 exposes the perfectly-regular request intervals and absence of mouse/keyboard events that AI Agents physically cannot replicate.

L1
Network · TLS Fingerprint JA3 / JA4 Fingerprint

Captures TLS handshake characteristics to precisely distinguish client types: Chrome browser / Python requests / curl / Cursor IDE. Spoofing requires reimplementing the full TLS protocol stack — the highest-cost evasion method.

L2
Application · HTTP Fingerprint User-Agent · Header Combination

Collects User-Agent, Accept-Language, Accept-Encoding, and custom headers. Official AI tools rarely actively spoof UAs; header order and combination characteristics expose impostors.

L3
Device · Browser Fingerprint Canvas · WebGL · Font List

Collects Canvas render hash, WebGL hash, font list, resolution, and timezone. Headless Chrome typically has missing Canvas, zero-value WebGL hash — a strong automation signal. Spoofing requires simulating a full GPU rendering context.

L4
Behavior · HDP² Fingerprint Mouse Trajectory · Keystroke Rhythm · Request Interval

Captures mouse trajectories, keystroke rhythms, and request interval mean/std deviation. A request interval std deviation <50ms triggers a "regular request" signal — human behavior that AI Agents physically cannot fully simulate.

L5
Identity Association Layer OAuth Token · Account–IP–Device Triple

Correlates employee login sessions within 5 minutes of the same IP, historical sessions within 1 hour from the same device fingerprint, and OAuth tokens — tracing every non-human access back to a responsible human. Auto-binds at confidence ≥80%; traces accountability with 95%+ accuracy.

>98% Compliant AI Tool
Detection Accuracy
>97% External Attack
Block Rate
<2% Compliant Tool
False Positive Rate
<2ms Added Latency
Per Request

Typical Risk Scenarios

Real-World NHI Attack Scenarios

Non-human identities typically carry high privileges, exist in large numbers, and lack monitoring — making them APT attackers' preferred entry point for bypassing traditional IAM defenses.

AI Assistant Overreach · Dev Scenario

Cursor AI Accesses Production Database

A developer asks Cursor to "check the user table in the production database." Cursor immediately uses the developer's OAuth token to access production — traditional IAM sees only "developer normal login," completely blind to AI being the actual operator.

ITDR Response: 5-layer fingerprint identifies Cursor behavior · links responsible developer · app authorization matrix blocks production DB access
Authorized Agent Out-of-Scope · Finance Scenario

AI Agent Aggregates Cross-System Sensitive Data

A finance user authorizes OpenClaw AI to run monthly reports. The Agent auto-logs into CRM to pull customer data, then cross-accesses ERP invoice data — OAuth scope is violated, but the IAM system cannot detect it.

ITDR Response: Identifies OpenClaw greylist · links employee's OAuth token · pushes approval request via DingTalk/Slack
External AI Attack · Credential Stuffing

Browser Use Automated Credential Stuffing

An attacker uses the Python library browser-use to simulate login attempts. WAF sees legitimate HTTPS traffic; credentials are real (obtained via stuffing). 4,912 requests are completely invisible to traditional defenses.

ITDR Response: TLS + behavioral fingerprint hits blacklist · all 4,912 requests blocked · IP range auto-banned

Core Value

Core Value

Why Is NHI Governance the Key to Zero Trust?

Non-human identities typically outnumber human accounts by 10× or more — representing the largest attack surface expansion in modern enterprises. Comprehensive NHI governance is the prerequisite for Zero Trust architecture to truly take effect.

Eliminate NHI Blind Spots

From API keys to AI Agents — full automated discovery builds a single trusted NHI asset ledger, making every machine identity queryable, manageable, and controllable.

Automated Least Privilege

No more manual security reviews. Behavior-driven analysis automatically recommends least-privilege policies — from concept to practice, significantly reducing lateral movement and privilege escalation risk.

Full Compliance Support

NHI asset ledgers, access logs, and privilege change audits are fully retained — meeting MLPS 2.0, ISO 27001, and industry regulatory requirements for identity and access management compliance.


vs. Traditional Security

How WuThreat ITDR Differs

Traditional products ask "is the account legitimate?" — we ask "is a real human operating it?" This solution complements existing IAM / WAF / DLP / SIEM, filling the gap in non-human identity subject determination.

Capability Traditional IAM Traditional WAF DLP EDR WuThreat ITDR Gateway
Identify AI Agent Tool Type
Endpoint visible
✓ 12+ mainstream tools
Distinguish AI Agent from Human
Simple rules only

Endpoint visible
✓ 5-layer fingerprint verdict
Attribute to Responsible Human
Account-level only
✓ 95%+ accuracy
Block AI Agent Unauthorized Access
Coarse-grained

Endpoint blocking
✓ App authorization matrix
Zero Business System Changes
Proxy needed

Agent install needed
✓ Gateway bypass, zero-intrusion
Cloud SaaS AI Support (e.g., Operator)
Traffic visible, not identified

No endpoint process
✓ 5-layer full coverage

Frequently Asked Questions

Common Questions Answered

We already have IAM and WAF. Why do we need this?

IAM and WAF verify "account legitimacy" and "traffic legitimacy." But in the AI Agent era, a legitimate account + legitimate traffic ≠ a human operating it. This solution fills the "identity subject determination" layer — answering "is this session a human or an AI?" We are complementary to IAM/WAF, not a replacement.

Will it false-positive on our compliant enterprise RPA bots?

No. We designed a "linked natural person session" as a -50% negative weight factor — as long as the RPA can be linked to an employee or department, the confidence score drops immediately. Compliant RPA false positive rate is <1%. One-click whitelisting is also supported (no alerts, but audit logs are still recorded).

How do you handle employee-authorized AI Agents, like using AI to run reports?

This is a typical "requires review" scenario. We: ① Identify the AI tool (greylist) → ② Link the employee identity (natural person session) → ③ Trigger DingTalk/Slack approval → employee confirms or the request is blocked. Employees can still use AI productively without allowing unauthorized AI operations.

Does deployment affect business systems or require code changes?

No business system changes required. The gateway deploys at the traffic entry point (inline or bypass mirror), passively observing TLS/HTTP characteristics. No client/agent installation, no SDK modification, no DNS changes. Bypass mode adds zero latency; gateway mode adds <2ms per request. Standard deployment completes in 3–5 business days.


Expert Consultation

Dedicated Identity Security Strategy Consultation

Facing complex identity threats (ITDR) and compliance challenges, you need customized solutions. Contact the WuThreat Identity Security expert team immediately, and we will provide you with tailored services based on your business architecture and industry characteristics:

Customized Practical Solutions

Tailored to your business scenarios

Emerging Threat Strategic Planning

Forward-looking long-term security strategy

Professional Technical Support

Expert team providing full technical assistance

Security Expert 1
Security Expert
Security Expert 2
Security Expert

Scan WeChat QR Code to Consult

Services Hotline 010-80716066
Business Email services@wuthreat.com