Lakera Guard focuses on detecting prompt injection at the input layer. RaksHex operates one layer deeper — authorizing the action the model wants to take, independent of how the prompt was worded, and enforcing that decision at the credential.
| Feature | Lakera Guard | RaksHex |
|---|---|---|
| Prompt Injection Detection | Focused, research-backed detection API | Not the primary focus — operates one layer deeper, at the action |
| Semantic Action Authorization | Not available | Evaluates each action against delegated authority before it runs |
| Delegated Authority & Attenuation | Not available | Child scope can't exceed parent scope, enforced in code |
| Credential Mediation | Detection only, no credential control | DENY blocks the credential itself, fail-closed |
| Tamper-Evident Action Ledger | Not available | Hash-chained record of every authorization decision |
| Policy Engine | Not available | Priority-ordered rules, with shadow-mode policy testing |
| MCP Tool Governance | Not available | Adversarial-intent scanning on MCP tool calls |
| PII Detection & Redaction | Available | Not a primary focus of the current product |
| Compliance Evidence | Not available | SOC 2 audit in progress; ledger data exportable as evidence |
| Self-hosted Option | Enterprise only | Available for self-hosted deployments |