The Trust Layer
The discipline that keeps the system honest. Memories are not accepted automatically; they are reviewed, curated, and approved as consented, traceable evidence before they ever inform a live representation.
We are not building the most convincing imitation of a human being. We are building the most trustworthy one.
Human curators remain in the loop at the decisions that matter. AI supports judgment; it does not replace it.
Many AI memory systems ingest whatever content is available and treat every source as equally credible. Forever Spoken's governance model does not. Each memory passes through review for provenance, ownership, and corroboration before it can inform a live representation.
Roles
The Subject
The person whose identity the persona reflects. Their consent and confirmation carry the highest authority in the model.
The Contributor
Family, colleagues, and communities who provide evidence. Contributors retain ownership of the memories they supply.
The Curator
Reviews evidence, resolves conflicts, and sets verification status. Uncertainty is flagged for steward review rather than resolved by assumption.
Governance Rules
Governance and the Claim Model
The framework treats an AI response as a collection of material claims, each weighed on its evidence provenance, source authority, corroboration, temporal relevance, perspective status, uncertainty, and disclosure authorization. Governance is the layer that resolves conflicting perspectives and decides what a persona is authorized to disclose. When the evidence does not support a claim, abstention is a governance decision, not a model accident.
Provenance-First Memory Architecture
Persona conclusions are deliberately kept separate from the contributor-owned evidence they rest on. Four layers keep synthesis and source distinct, so every trait remains traceable and never fabricated.
Layer 1
Synthesized, high-level conclusions about the person. Held separately, with no contributor ownership transferred into it.
Layer 2
Who supplied what, and their ownership and relationship to the subject. Each contributor keeps rights to their memories.
Layer 3
The discrete, source-attributed memories: the traceable evidence to which every persona trait is anchored.
Layer 4
Links each synthesized trait back to the specific evidence that supports it, so conclusions are always accountable.
Interactive Illustration
A working illustration of how curators review pending evidence, inspect provenance metadata, weigh fidelity, and verify a Memory Evidence Object into the persona. Sample data shown for demonstration.
// Standards Alignment
Our governance is explainable, auditable, and transparent by design, and is aligned with the NIST AI Risk Management Framework: Govern at the center, with Map, Measure, and Manage around it.
See the open questions we are working to answer about trust across generations.
