Frequently askedquestions.
PRE-GEN is an open protocol for pre-generation authorization in generative AI. It answers “is this content authorized?” before the model produces anything — not after. A rights registry, a verification API, two-party cryptographic signatures, and a hash-chained audit log give rights holders control over how their identity, brand, voice, style, or creative work is used in AI-generated content — and give providers signed, storable proof of compliance.
Provenance standards like C2PA describe what was generated. PRE-GEN determines whether it should be generated. They are complementary: PRE-GEN can export a signed assertion that embeds into a C2PA manifest, but the authorization decision happens before generation, not after.
No — and we say so plainly. PRE-GEN standardizes compliant generation flows. It cannot prevent unauthorized generation by uncooperative providers, local open-source inference, or offshore services. Its value is proportional to the size of the compliant ecosystem: the more providers that integrate, the stronger the protection and the stronger the market norm of asking first.
The opt-out is the strongest protection in the protocol. It is checked before any license is consulted, and no license — not even a premium one with co-signed exceptions — can override it. A subject who opts out of a category is invisible to compliant generation in that category, full stop. Registration and the opt-out are free.
The permanent identifier every subject and license carries — e.g. PG-000042* for a subject, PG-STD-000001-K7M2QX9 for a license. Crockford base32 (no I/L/O/U) with a mod-37 check character that catches every single-character substitution and adjacent transposition — designed to be retyped correctly from a screenshot. The format is frozen; codes are cited in licenses, disputes, and court filings. See the full specification.
Today, a single registry operates the verification API and issues credentials — PRAMPTA, the protocol’s first commercial implementation. The specification itself is open (Apache-2.0) with published byte-level test vectors, so anyone can implement it; the format explicitly reserves room for additional issuers. The long-term intent is a federated, multi-operator network — and the protocol is designed so codes issued today remain valid in that future.
Yes. The crypto invariant, the PG code format, versioning rules, and the full refusal-code registry are published on this site (start here) and in the SDK repository under Apache-2.0, with byte-level test vectors that pin every implementation. A formal conformance suite is on the roadmap; today the vectors are the conformance contract.
The signed refusal and the audit chain are evidence: they establish that the provider was told “no”, by whom, when, and under which rule version — independently verifiable by any third party. PRE-GEN does not replace law enforcement or courts; it gives them something verifiable to work with.
The protocol, SDKs, and specification are free. A few lines of code before your model runs. Provider credentials are issued through prampta.com; a sandbox tier verifies against synthetic subjects only, so integration costs engineering time, nothing else.
Decisions reference the operator key fingerprint (pg-ed25519:…) they were signed with, and the key set is published at GET /keys with rotation-awareness: old keys remain available for verification of past decisions, new decisions use the new key. Stored evidence bundles remain verifiable offline, years later.