AI REQUESTS↯
0Last 24 hours metadata only
ON-PREMISE?
AI SECURITY POSTURE
Good morning.
Connect the first application to begin measuring AI traffic.
Universal configuration foundationProviders, policies and metadata-only security events are managed per organization.
View scope ↗BLOCKED×
0Policy decisions no data yet
REDACTED⌁
0Sensitive values no data yet
CONNECTED APPS⇄
0Gateway clients ready to connect
CONTROLLED TRAFFIC PATH
READYGateway routing
01AIInternal appNot connected
02
CyberPassInspect · Decide · Route
CyberPassInspect · Decide · Route03A◉Private LLMLocal pilot provider
03B↗Public LLMNot configured
PROTECTION COVERAGE
DetailsBaseline controls
Gateway coreRegistry routing
Text DLPBaseline
LLM GuardIn development
Identity & RBACBaseline
Provider & policy registryBaseline
Coverage reflects implementation stage, not a security score.
SECURITY EVENTS
Latest decisions
No events recorded yet.
GET STARTED
0 / 4Deployment path
LIVE AI TRAFFIC
Gateway decisions
Metadata-only view of recent AI requests. Prompt and response content is never displayed.
Polling every 5s
Waiting for Gateway traffic.
INVESTIGATION DESK
Security events
Filter, classify and export metadata evidence for Security Operations.
No matching events.
ENTERPRISE CONNECTIONS
Identity, trust & SIEM
Configure reusable protocol adapters once per organization. Customer URLs, certificates, credentials and role mappings stay in tenant-scoped records.
OIDC / LDAP / ACTIVE DIRECTORY
Identity providers
No enterprise identity provider configured.
CUSTOMER PKI
Trust bundles
System trust store only.
CREDENTIAL PROTECTION
CHECKINGSecret storage
Checking encryption configuration.
CEF / SYSLOG
SIEM destinations
No SIEM destination configured.
CONTROLLED TOOL EXECUTION
MCP Firewall
Register Streamable HTTP servers once, expose only approved tools and require analyst approval for sensitive actions.
STREAMABLE HTTP
TENANT SCOPEDServer registry
No MCP servers configured.
TOOL ALLOWLIST
Policies
No tool policies configured.
HUMAN IN THE LOOP
Pending approvals
No pending approvals.
METADATA ONLY
Recent tool calls
No MCP traffic recorded.
Arguments and tool outputs are inspected in transit but are not stored. Audit records keep names, decisions and SHA-256 argument fingerprints.
UNIVERSAL LLM CONTROL
Providers, models & routing
Connect compatible inference endpoints once, then expose stable aliases to every approved application.
PROVIDER REGISTRY
Inference endpoints
No providers configured.
MODEL CATALOG
Stable model aliases
No model aliases configured.
VERSIONED ENFORCEMENT
AI security policies
Create an immutable draft, simulate it with sample text, then activate it for Gateway traffic.
POLICY REGISTRY
Versions
No policy versions configured.
DATA PROTECTION
DLP detectors
Built-in Kazakhstan and secret detectors plus organization-specific dictionaries and safe regular expressions.
DETECTOR REGISTRY
Active inspection rules
Loading detectors.
WORKSTATION PROTECTION
Endpoint devices
Windows-first local Connector for AI agents, IDE tools and the managed Chrome/Edge extension.
DEVICE REGISTRY
LOOPBACK PROXYManaged connectors
No endpoint devices enrolled.
Connector listens on 127.0.0.1 only. Browser protection inspects text and supported files before submission; direct egress controls remain a customer firewall/MDM responsibility.
ENROLLMENT TOKENS
ONE-TIMEPending installations
No enrollment tokens created.
IDENTITY FOUNDATION
Users & application access
Human access is role-based. Applications authenticate with scoped, revocable keys.
HUMAN ACCESS
Organization members
Sign in to load members.
APPLICATION ACCESS
Service accounts
Sign in to load service accounts.
EXPLAINABLE AI THREAT SIGNALS
LLM Guard
Detect suspicious instruction patterns, combine them into a risk score and apply the active policy threshold.
Prompt injection
EXPLAINABLESignals for instruction override, hidden context requests and role confusion.
Exfiltration
TEXTSignals for secret disclosure, system-prompt extraction and credential requests.
Tool coercion
POLICYSignals that attempt to force commands or unsafe tool behavior through text.
DECISION MODEL
Signal → score → action
Guard does not promise perfect detection. Every matched signal is recorded with the request metadata so an analyst can understand why CyberPass observed or blocked a request.
ObserveReview thresholdBlock threshold
SAFE VALIDATION
Test lab
Validate text, file and policy behavior with synthetic data. Tests do not call an upstream model.
DLP inspection
Check built-in and organization detectors, see matched rules and preview redaction.
File inspection
Extract and inspect a supported synthetic document with the active file policy. Contents are processed in memory and are not stored.
Policy simulation
Test a specific immutable policy version and model alias before activation.
Use synthetic data only.Do not paste real customer, payment or production secrets into the lab.
ON-PREMISE OPERATIONS
Deployment health
Runtime component state and the minimum readiness gates for a controlled pilot.
RUNTIME
just nowComponents
Control APIConnecting
—GatewayRuntime available
READYPublic providerNo external route
OFFEvent storeMetadata only
READYACCEPTANCE
CHECKSPilot readiness
Checking pilot readiness.
ORGANIZATION SETTINGS
Security data
Control the retention period for minimized security-event metadata.
Event retention
Prompt and response content is not stored in the current metadata-only mode.