Threat Models¶

This document provides platform-level threat modeling for ConnectSoft's core platforms and services using STRIDE methodology. It is written for security architects, platform engineers, and developers who need to understand threats and mitigations across the ConnectSoft ecosystem.

Threat modeling at ConnectSoft follows a systematic approach, analyzing threats across all STRIDE categories and including AI-specific threats. Each platform and SaaS product can reference these threat models as baselines, adding product-specific details as needed.

Important

Threat Modeling is Continuous: Threat models are living documents that should be updated as systems evolve, new threats emerge, and mitigations are implemented. This document provides baseline threat models that should be refined per product and reviewed regularly.

Methodology¶

STRIDE Categories¶

STRIDE is a threat modeling methodology that categorizes threats into six categories:

Spoofing:

Impersonating another user, service, or system
Examples: Stolen credentials, token theft, OAuth misuse, service impersonation

Tampering:

Unauthorized modification of data or code
Examples: Token tampering, configuration tampering, data modification, code injection

Repudiation:

Denying that an action occurred
Examples: Lack of audit logs, missing authentication records, untraceable actions

Information Disclosure:

Unauthorized access to information
Examples: Token leaks, PII exposure, data breaches, log exposure

Denial of Service (DoS):

Preventing legitimate users from accessing services
Examples: Brute force attacks, resource exhaustion, API flooding, DDoS

Elevation of Privilege:

Gaining unauthorized access or privileges
Examples: Privilege escalation, mis-scoped roles, unauthorized access to admin functions

AI-Specific Threats¶

In addition to standard STRIDE threats, AI systems face unique threats:

Prompt Injection:

Malicious input designed to manipulate AI model behavior
Examples: User input containing injection attempts, malicious documents, adversarial prompts

Data Exfiltration Through Tools:

AI agents using tools to exfiltrate data across tenant boundaries
Examples: Cross-tenant data access via shared tools, unauthorized API calls, context leakage

Model Misuse:

Using AI models for unintended purposes
Examples: Jailbreaks, bypassing safety filters, generating harmful content

Jailbreaks:

Techniques to bypass AI model safety controls
Examples: Prompt engineering to bypass filters, adversarial inputs, model manipulation

See: Security Overview for security principles.

See: Patterns Cookbook for mitigation patterns.

Threat Model: Identity & Auth¶

Assets¶

User Identities:

User accounts, credentials, profiles
Multi-factor authentication (MFA) settings
External identity provider links

Tokens:

Access tokens (short-lived)
Refresh tokens (long-lived, rotated)
ID tokens (user identity claims)

Credentials:

Passwords (hashed, never stored in plain text)
MFA secrets (TOTP seeds, recovery codes)
API keys and client secrets

Data Flow¶

The following diagram illustrates the identity and authentication data flow with trust boundaries:

flowchart TD
    A[User/SaaS Frontend] -->|Auth Request| B[API Gateway]
    B -->|Validate & Route| C[Identity Platform]

    C -->|Validate Credentials| D{External IdP?}
    D -->|Yes| E[External IdP<br/>Azure AD/Google]
    D -->|No| F[Internal User Store]

    E -->|User Claims| C
    F -->|User Data| C

    C -->|Issue Tokens| B
    B -->|Return Tokens| A

    A -->|API Request with Token| G[Service API]
    G -->|Validate Token| C
    C -->|Token Valid| G
    G -->|Process Request| H[Business Logic]

    C -->|Audit Events| I[Audit Platform]
    G -->|Audit Events| I

    style C fill:#e1f5ff
    style E fill:#fff4e1
    style F fill:#fff4e1
    style I fill:#e8f5e9

Hold "Alt" / "Option" to enable pan & zoom

Trust Boundaries:

External Boundary - User/SaaS Frontend to API Gateway (public internet)
Internal Boundary - API Gateway to Identity Platform (internal network)
External IdP Boundary - Identity Platform to External IdP (external network)
Service Boundary - Service API to Identity Platform (internal network)

Threats by STRIDE Category¶

Spoofing¶

Threats:

Stolen Tokens - Attacker steals access or refresh tokens
OAuth Misuse - Attacker uses compromised OAuth client credentials
Credential Theft - Attacker steals user passwords or MFA secrets
Service Impersonation - Attacker impersonates Identity Platform or downstream services

Mitigations:

Short-lived access tokens (15-60 minutes)
Refresh token rotation on use
Token binding (bind tokens to client IP/user agent)
Strong password policies and MFA enforcement
Service-to-service authentication using managed identities
Certificate pinning and token signature validation

Tampering¶

Threats:

Token Tampering - Attacker modifies token claims or signatures
Configuration Tampering - Attacker modifies identity provider configuration
User Data Tampering - Attacker modifies user profile or permissions
Audit Log Tampering - Attacker modifies or deletes audit logs

Mitigations:

Cryptographically signed tokens (JWT with RSA/ECDSA)
Token signature validation on every request
Configuration stored in secure, version-controlled config store
Role-based access control (RBAC) for configuration changes
Immutable audit logs with integrity checks
Regular configuration audits

Repudiation¶

Threats:

Lack of Login Audit - No record of authentication attempts
Missing Token Usage Logs - No record of token validation or usage
Untraceable Admin Actions - Admin actions not logged or attributed

Mitigations:

Comprehensive audit logging of all authentication events
Token validation and usage logged with correlation IDs
All admin actions logged with user identity and timestamp
Audit logs stored in immutable, tamper-evident store
Regular audit log reviews and compliance reporting

Information Disclosure¶

Threats:

Token Leaks - Tokens exposed in logs, URLs, or error messages
PII Exposure - User PII exposed in logs or API responses
Credential Exposure - Passwords or secrets exposed in logs or config
Configuration Exposure - Identity provider configuration exposed

Mitigations:

Tokens never logged or included in URLs
PII redaction in logs and error messages
Secrets stored in Key Vault, never in code or config
Configuration access restricted to authorized personnel
Data classification and handling policies
Regular security scanning for exposed secrets

Denial of Service¶

Threats:

Brute Force Attacks - Repeated login attempts to guess passwords
Token Validation Flooding - Excessive token validation requests
Auth API Flooding - DDoS attacks on authentication endpoints
Resource Exhaustion - Exhausting identity service resources

Mitigations:

Rate limiting on authentication endpoints
Account lockout after failed attempts
CAPTCHA or challenge-response for suspicious activity
Token validation caching and rate limiting
DDoS protection at network and application layers
Resource quotas and throttling

Elevation of Privilege¶

Threats:

Privilege Escalation - User gains unauthorized privileges
Mis-Scoped Roles - Tokens contain broader scopes than intended
Admin Function Access - Unauthorized access to admin functions
Cross-Tenant Access - User accesses resources from another tenant

Mitigations:

Principle of least privilege in role definitions
Scope-based authorization (tokens scoped to specific resources)
Tenant-scoped access control (all resources tagged with tenant ID)
Regular access reviews and role audits
Break-glass procedures for elevated access
Multi-factor authentication for admin functions

See: Identity Platform for platform details.

See: Patterns Cookbook for implementation patterns.

Threat Model: Billing & Payments¶

Assets¶

Payment Data:

Payment service provider (PSP) tokens
Payment method references (never full card numbers)
Billing addresses and payment metadata

Subscription State:

Subscription status and lifecycle
Plan and edition assignments
Usage limits and quotas

Billing Events:

Metering events (usage tracking)
Invoice generation and delivery
Payment processing events

Financial Records:

Invoices and billing history
Payment transactions
Refund and credit records

Threats¶

Fraudulent Usage:

Attacker creates fake accounts to consume free tier resources
Attacker manipulates usage metering to avoid charges
Attacker uses stolen payment methods

Payment Tampering:

Attacker modifies invoice amounts
Attacker manipulates payment processing
Attacker modifies subscription state to avoid charges

Metering Tampering:

Attacker modifies usage metering events
Attacker manipulates usage calculations
Attacker bypasses usage limits

Invoice Fraud:

Attacker generates fraudulent invoices
Attacker modifies invoice data
Attacker manipulates billing cycles

Mitigations¶

Payment Service Provider (PSP) Usage:

Never store full payment card data
Use PCI DSS-compliant PSP (Stripe, PayPal, etc.)
Payment data only in PSP tokenized form
PSP handles all payment card processing

Signed Metering Events:

All metering events cryptographically signed
Event integrity validated before billing
Immutable event log for audit trail
Double-entry style ledger for financial records

Restricted Roles:

Billing and payment operations require elevated roles
Multi-factor authentication for payment operations
Separation of duties (billing admin vs. payment processor)
Regular access reviews for billing roles

Audit and Compliance:

All billing events logged to Audit Platform
Financial records stored in tamper-evident store
Regular financial audits and reconciliation
Compliance with financial regulations

Rate Limiting and Fraud Detection:

Rate limiting on subscription creation
Fraud detection for suspicious payment patterns
Account verification for high-value subscriptions
Usage anomaly detection

See: Billing & Subscription Platform for platform details.

Threat Model: Documents & Forms¶

Assets¶

Uploaded Documents:

User-uploaded files (PDFs, images, documents)
Document metadata and classifications
Document versions and history

Generated Documents:

System-generated documents (invoices, reports, forms)
Document templates and content
Document rendering and formatting

Form Submissions:

Form data and responses
File attachments to forms
Form metadata and workflow state

Document Storage:

Blob storage (encrypted at rest)
Database records (document metadata)
Backup and archive storage

Threats¶

Unauthorized Access:

Attacker accesses documents from another tenant
Attacker bypasses access control to view sensitive documents
Attacker accesses documents through API vulnerabilities

Tampering:

Attacker modifies stored documents
Attacker manipulates document metadata
Attacker tampers with form submissions

Injection via Form Content:

Malicious content in form submissions
File upload attacks (malicious files, oversized files)
Content injection in document processing

Data Leakage:

PII/PHI exposure in logs
Document content in error messages
Backup exposure or improper disposal
Cross-tenant data leakage

Mitigations¶

Data Classification:

Documents classified by sensitivity (Public, Internal, Confidential, Sensitive, Highly Sensitive)
Classification determines encryption and access control requirements
Automatic classification based on content analysis

Encryption:

Encryption at rest for all document storage
Encryption in transit for all document transfers
Key management via Key Vault
Per-tenant encryption keys where required

Row-Level Security (RLS):

Tenant-scoped access control at database level
Document access requires valid tenant context
RLS policies enforce tenant isolation

Redaction:

PII/PHI redaction in logs and error messages
Document content redaction for lower-privilege access
Safe logging practices (no document content in logs)

Input Validation:

File type validation and restrictions
File size limits and quotas
Content scanning for malicious files
Form input sanitization and validation

Integrity Checks:

Document hashes stored for integrity verification
Tamper detection through hash validation
Version control with integrity checks
Audit trail for all document access and modifications

See: Documents Platform for platform details.

See: Patterns Cookbook for implementation patterns.

Threat Model: Integrations (Webhooks & Connectors)¶

Assets¶

Webhook Endpoints:

Inbound webhook URLs and endpoints
Webhook payloads and event data
Webhook delivery status and retry logic

Connector Credentials:

External system API keys and tokens
OAuth tokens for external services
Per-tenant connector configurations

Inbound/Outbound Data:

Data received from external systems
Data sent to external systems
Data transformation and mapping

Threats¶

Spoofed Webhooks:

Attacker sends fake webhook events
Attacker replays old webhook events
Attacker manipulates webhook payloads

Replay Attacks:

Attacker replays valid webhook events
Attacker replays connector API calls
Time-based replay attacks

Connector Abuse:

Attacker uses connector to pivot into external systems
Attacker abuses connector permissions
Attacker exfiltrates data through connectors

Exfiltration Through Connectors:

Attacker uses connector to send data to unauthorized destinations
Attacker manipulates connector configuration
Cross-tenant data exfiltration via shared connectors

Mitigations¶

Webhook Signatures:

All webhooks signed with shared secrets or HMAC
Signature validation on every webhook request
Per-tenant webhook secrets stored in Key Vault
Signature algorithm and secret rotation

Replay Protection:

Nonce or timestamp validation for webhook events
Event ID tracking to prevent duplicate processing
Time-window validation (reject events outside time window)
Idempotency keys for connector operations

IP Allowlists:

Webhook endpoints restricted to known source IPs
Connector outbound calls from known IP ranges
Network-level filtering and validation

Per-Tenant Key Vault Secrets:

Each tenant's connector credentials stored separately
Credentials scoped to tenant and specific external system
Credential rotation without affecting other tenants
Audit logging of credential access and usage

Strict Scopes:

Connector OAuth tokens scoped to minimum necessary permissions
Per-connector permission boundaries
Regular scope reviews and audits
Principle of least privilege for connector permissions

Throttling and Rate Limiting:

Rate limiting on webhook endpoints
Throttling of connector API calls
Quotas per tenant and per connector
Anomaly detection for unusual connector activity

See: Integration Platform for platform details.

See: Webhooks and Events for integration patterns.

Threat Model: AI Agents & Tools¶

Assets¶

Prompts:

User prompts and instructions
System prompts and guardrails
Prompt templates and configurations

Context:

Tenant data and context
Knowledge base content
Vector embeddings and retrieval results

Agent Configurations:

Agent definitions and capabilities
Tool access permissions
Agent execution policies

Tool Invocations:

API calls made by agents
Database queries executed by agents
External system interactions

Threats¶

Prompt Injection:

Malicious user input designed to manipulate agent behavior
Injection attempts in uploaded documents
Adversarial prompts to bypass safety controls

Data Exfiltration Across Tenants:

Agent accessing data from wrong tenant
Shared tools or vectors leaking tenant data
Context contamination between tenants

High-Privilege Tool Misuse:

Agent using high-privilege tools without proper authorization
Agent making unauthorized API calls
Agent performing destructive or sensitive operations

Model Misuse and Jailbreaks:

Techniques to bypass AI model safety filters
Generating harmful or inappropriate content
Manipulating model behavior for unintended purposes

Mitigations¶

Tool-Level Authorization:

Each tool requires explicit authorization per agent
Tenant-scoped tool access (tools can only access tenant's data)
Role-based tool permissions (different tools for different agent roles)
Tool invocation logging and audit

Hard Separation of Tenant Contexts:

Tenant context strictly enforced in retrieval and tools
No cross-tenant data access in RAG or vector stores
Tenant ID validated before any data access
Explicit "no cross-tenant context" rules in agent execution

Guardrails and Content Filters:

Input validation and sanitization for prompts
Content filters for generated responses
Instruction hierarchy (system instructions override user input)
RAG scoping by tenant (retrieval only from tenant's knowledge base)

Redaction & Classification:

Sensitive data redacted from prompts and logs
Data classification determines what can be included in prompts
PII/PHI redaction before sending to AI models
Safe logging practices (no sensitive data in agent logs)

Human-in-the-Loop:

Critical actions require human approval (payments, destructive changes)
High-privilege tool invocations require confirmation
Sensitive operations logged and reviewed
Escalation procedures for suspicious agent behavior

Prompt Security:

Prompt injection detection and prevention
Input sanitization and validation
System prompt protection (system prompts cannot be overridden by user input)
Adversarial prompt detection and blocking

The following diagram illustrates AI agent execution with trust boundaries:

flowchart TD
    A[User Input] -->|Prompt| B[AI Gateway]
    B -->|Validate & Classify| C[Content Filter]
    C -->|Sanitized Prompt| D[Agent Orchestrator]

    D -->|Tenant Context| E[Tenant Isolation Layer]
    E -->|Scoped Context| F[AI Model]

    F -->|Response| G[Tool Authorization]
    G -->|Authorized Tools| H{Tool Type?}

    H -->|High-Privilege| I[Human Approval]
    H -->|Standard| J[Tool Execution]

    I -->|Approved| J
    J -->|Tool Result| K[Response Assembly]

    K -->|Final Response| L[Content Filter]
    L -->|Validated| B
    B -->|Response| A

    D -->|Audit Events| M[Audit Platform]
    G -->|Audit Events| M
    J -->|Audit Events| M

    E -->|Tenant-Scoped RAG| N[Knowledge Base]
    N -->|Tenant Data Only| E

    style E fill:#e1f5ff
    style G fill:#fff4e1
    style I fill:#ffebee
    style M fill:#e8f5e9

Hold "Alt" / "Option" to enable pan & zoom

Trust Boundaries:

User Boundary - User Input to AI Gateway (public interface)
Content Filter Boundary - Input/output validation and filtering
Tenant Isolation Boundary - Strict tenant context enforcement
Tool Authorization Boundary - Tool access control and approval
Human Approval Boundary - Critical operations requiring human review

See: Factory Overview for Factory architecture.

See: Patterns Cookbook for AI security patterns.

Summary¶

Using These Threat Models¶

These threat models provide baseline threat analysis for ConnectSoft's core platforms. Each platform and SaaS product should:

Reference Baseline Models - Use these threat models as starting points
Add Product-Specific Threats - Identify threats specific to the product
Refine Mitigations - Adapt mitigations to product architecture
Regular Reviews - Update threat models as systems evolve

Platform-Specific Extensions¶

Identity Platform:

Reference Identity & Auth threat model
Add product-specific threats (external IdP federation, multi-tenant user management)

Billing Platform:

Reference Billing & Payments threat model
Add product-specific threats (usage metering, subscription lifecycle)

Documents Platform:

Reference Documents & Forms threat model
Add product-specific threats (document processing, version control)

Integration Platform:

Reference Integrations threat model
Add product-specific threats (connector marketplace, integration templates)

AI Factory:

Reference AI Agents & Tools threat model
Add product-specific threats (code generation security, agent collaboration)

Continuous Improvement¶

Threat models should be:

Reviewed Regularly - Updated as systems evolve and new threats emerge
Validated - Tested through security assessments and penetration testing
Documented - Changes tracked and communicated to stakeholders
Integrated - Part of design reviews and security architecture decisions

See: Security Overview for security principles.

See: Patterns Cookbook for mitigation implementation.

See: Compliance Blueprints for compliance considerations.

Security Documentation¶

Security Overview - Security vision and principles
Patterns Cookbook - Security implementation patterns
Compliance Blueprints - Compliance-by-design architecture

Platform Documentation¶

Identity Platform - Identity & Access Platform
Billing & Subscription Platform - Billing Platform
Documents Platform - Documents Platform
Integration Platform - Integration Platform
Audit Platform - Audit Platform

Factory & AI¶

Factory Overview - AI Factory architecture
Agent System Overview - Agent system

Integration¶

Webhooks and Events - Webhook patterns

Threat Models¶

Methodology¶

STRIDE Categories¶

AI-Specific Threats¶

Threat Model: Identity & Auth¶

Assets¶

Data Flow¶

Threats by STRIDE Category¶

Spoofing¶

Tampering¶

Repudiation¶

Information Disclosure¶

Denial of Service¶

Elevation of Privilege¶

Threat Model: Billing & Payments¶

Assets¶

Threats¶

Mitigations¶

Threat Model: Documents & Forms¶

Assets¶

Threats¶

Mitigations¶

Threat Model: Integrations (Webhooks & Connectors)¶

Assets¶

Threats¶

Mitigations¶

Threat Model: AI Agents & Tools¶

Assets¶

Threats¶

Mitigations¶

Summary¶

Using These Threat Models¶

Platform-Specific Extensions¶

Continuous Improvement¶

Related Documents¶

Security Documentation¶

Platform Documentation¶

Factory & AI¶

Integration¶