Threat Model

SentinelRN is a risk-based security framework. It does not claim perfect protection — mobile runtime checks can be bypassed, prompt-injection detection can miss attacks, and PII detection can produce false positives and negatives. The goal is to help developers make better, more informed security decisions.

Threats SentinelRN addresses

1. Compromised runtime environments — rooted, jailbroken, emulated, or otherwise modified devices that weaken platform protections.
2. Runtime instrumentation & tampering — debuggers, Frida, hooking frameworks, method swizzling, repackaged builds.
3. Unsafe AI input handling — secrets, PII, or injection sent to an AI provider without inspection.
4. Sensitive data leakage — secrets/PII leaking into logs, analytics, crash reports, or AI payloads.
5. Poor application response to risk — boolean-only APIs with no severity, confidence, or recommended action.

What SentinelRN does not protect against

• A fully compromised backend
• A malicious authorized user
• Stolen credentials already accepted by the server
• A determined attacker with full device control
• All possible root/jailbreak bypasses or prompt-injection attacks
• Insecure API design or AI provider configuration

Backend still matters

SentinelRN should never be your only security boundary. Continue to enforce backend authorization, rate limiting, token revocation, server-side prompt validation, audit logging, and abuse detection.

The client can reduce risk. The server must enforce trust.

The full threat model lives in the repository: docs/THREAT_MODEL.md.