Full security model specification: how PromptImageLab achieves zero server-side data retention, AES-256 API key encryption, SOC 2 Type II readiness, and prompt injection defense layers.
Zero Server-Side Data Retention
No prompt text, AI model responses, or user-submitted content is stored on PromptImageLab servers. All inference calls are proxied directly from the client to the provider's API. Server logs contain only anonymized rate-limiting metadata with a 24-hour rolling TTL.
Client-Side API Key Encryption
API keys provided by users are encrypted using AES-256-GCM with a per-session derived key before storage in localStorage. Keys are never transmitted to PromptImageLab servers. On browser close, the session key is destroyed and the encrypted blob cannot be decrypted without the original session.
Prompt Injection Defense Layers
Studio's Prompt Evaluator runs a three-layer injection scan: (1) Static pattern matching against known injection payloads, (2) LLM-as-judge evaluation of whether the system prompt maintains boundary integrity after appending user input, and (3) XML/JSON output validation to detect schema-violating responses that indicate successful injection.