Trust policy
Trust policy
Verbitas produces tamper-evident provenance records. We explicitly do not:
- Claim content is “real”, “fake”, “authentic”, or “genuine”
- Assert semantic truth about content
- Promise legal compliance with any regulation
- Guarantee that provenance records cannot be fabricated by a determined adversary with sufficient resources
Verification states
Our verifier always returns one of the states in the closed enum. Every result always displays:
- What this proves — the specific cryptographic claim
- What this does not prove — semantic truth, content accuracy, legal compliance
Explainability requirement
No verification result is emitted without both proves and does_not_prove fields.
This is enforced at the API level and in all SDK response types.
Robustness claims
Watermark robustness claims are documented in Watermarking. Robustness benchmarks are run in CI and regressions past threshold fail the build.
Threat model
The threat model is maintained in docs/THREAT-MODEL.md. Every threat has at least
one corresponding negative test.