Comparison
One row per capability. Decide from the evidence.
Kevros operates at a different layer than content moderation or training-time alignment. The table below states the architectural differences without commentary on any vendor's broader product.
| Capability | Kevros | Training-time alignment | Post-hoc moderation API | Content classification service | Inline guardrail service |
|---|---|---|---|---|---|
| Enforcement model | Fail-closed cryptographic gate | Training-time RLHF | Post-hoc filter | Content classifier | Inline content filter |
| Evidence chain | Hash-chained, signed, verifiable | Unstructured (model weights) | Non-cryptographic (text logs) | Non-cryptographic (API logs) | Non-cryptographic (CloudWatch) |
| Post-quantum signing | ML-DSA-87 (FIPS 204) | None | None | None | None |
| Formally verified | TLA+, 1.94B states | None | None | None | None |
| Vendor lock-in | None (works with any provider) | Single provider | Single provider | Single cloud | Single cloud |
References for each Kevros row are on the Proof page.