Security Posture & Compliance

Threat Model

The application operates as a client-side tool. The primary trust boundary is the user's device.

API Keys: Stored in process.env. Must not be leaked in client bundles (Note: For this demo, keys are assumed to be handled by the build environment).
User PII: Email and names stored in LocalStorage.
Generated IP: Architectural artifacts generated by the tool.

Note: The current implementation is a simulation for demonstration purposes.

Algorithm: Simulates Argon2 hashing (currently btoa for demo).
Rate Limiting: Enforces a strict 5-attempt limit per 15-minute window to prevent brute-force attacks.
Session: LocalStorage-based session tokens.

Vulnerability	Mitigation Strategy
Injection	All LLM inputs are sanitized. `react-markdown` sanitizes HTML output.
Sensitive Data Exposure	`moderationService` automatically masks PII (Email, IP, Phone) before storage.
Broken Access Control	Governance gates restrict "High Risk" templates to specific roles during onboarding.
Logging & Monitoring	All moderation actions are recorded in an immutable audit log (`se_architect_moderation_audit`).

To prevent "Jailbreaking" or unsafe generation:

System Prompt Locking: The META_PROMPT is injected as a systemInstruction and cannot be overridden by user input.
Output Validation: Quality gates check for specific safety markers in the generated text.