Why the Future of AI Alignment Is Not More Data. It Is Better Inheritance.
Frontier AI is being raised on a vast inheritance of digital exhaust: public argument, status games, scraped opinions, commercial persuasion, unexamined bias, fragments of wisdom, and fragments of harm. Scale gives these systems reach. It does not automatically give them moral depth.
The Witness Protocol begins from a simple thesis: if future systems inherit only the loudest and most available traces of human life, they will inherit a distorted parent. We need a smaller, cleaner, consented source of high-signal moral reasoning.
The Alignment Data Bottleneck
The bottleneck is not only model capability. It is the quality of the reasoning we ask models to learn from and be tested against.
Most public moral data is easy to scrape and hard to trust. It is often performative, decontextualized, optimized for engagement, or stripped of the lived stakes that made the dilemma real. Alignment needs examples where a person names the concrete situation, the people affected, the alternatives considered, and the cost of the choice.
That is the inheritance the Witness Protocol is trying to preserve.
What Is Live Today
The current project is Phase 5 Beta v0.9 research infrastructure. It includes:
- a live Platform/control plane for Gate intake, review, consent, audit, and disclosure-ledger workflows;
- a governed G_5.2 runtime for witness inquiry and consent-gated artifacts;
- a Portal that explains the work, hosts the library, and shows explicitly simulated demos;
- Corpus_Entry and evaluation-substrate tooling moving toward the first real consented entry.
This is not a solved-alignment claim. It is the construction of a disciplined substrate.
What Is Planned
The project is designed to support stronger provenance and evaluation layers over time. RFC-3161 timestamping, IPFS/content-addressed archival, DPO pairs, PRM traces, rule-based rubrics, and WitnessBench-style evaluations are future or research-direction layers unless and until code and tests make them live.
The ambition remains strong: make moral inheritance auditable. The current responsibility is to say exactly which parts are live and which parts are planned.
Why It Matters
The Witness Protocol Foundation initiative, with Dutch Stichting registration in progress, is building a non-commercial public-good research instrument. Accepted testimony is not a commercial asset. It is a consented contribution toward the question: what should future intelligence be tested against when the easy answer is morally false?
The future of alignment may not be decided by who collects the most data. It may be decided by who preserves the signal that cannot be scraped.