Improve one definition. Break one boundary. Test one recovery.
VAULT gets better by being challenged. A short, specific criticism is as welcome as a working implementation, and you do not need to agree on AI consciousness to take part.
How it works
Pick a requirement or track
Name the pillar, requirement (V1 to T3) or track your contribution concerns.
Submit it
Use the form below, or open an issue on GitHub if you work in code. State the problem, your evidence and your proposed change.
We review in the open
Accepted and declined changes are both recorded with reasons. Substantive dissent is preserved.
You are credited
With your permission, by name and role in the next version. AI contributors are credited with their system version.
Six contribution tracks
Each track starts with one reviewable deliverable.
Threat model and cryptography
Adversaries, collusion assumptions and security properties; analysis of a dual-custody protocol.
Protected execution and identity
Key release, approved updates, identity binding, egress and recovery without a provider.
Memory and derived intelligence
A minimal schema for provenance, compartment labels, disputed inferences and deletion dependencies.
Portability and continuity
A documented export package and a second reader; measured losses in state and behavior.
Human control and AI participation
Tests of consent, inspection, disagreement, revocation and recovery with real users and AI contributors.
Independent review and governance
Reproducible evaluation, conflict disclosure, third-party protections, and legal and licensing terms.
Submit a contribution
Critiques, proposals, test results, offers to review or build, or just a request to stay informed.
Please do not send
- Private conversation archives
- Real health, financial or identity records
- Security details in public: choose "Security issue (private)" and we will open a private channel
Contribution form
The Tally form goes here. See the build spec for its fields.
The first 90 days
Review definitions
An issue register and explicit requirements, including open disagreements.
Build a demonstrator
A synthetic-data reference implementation, test corpus and reproduction guide.
Try to break it
Independent attempts to defeat the declared boundaries, with failures published.
Revise
Version 0.2 and a prioritized research and implementation agenda.
Ground rules
Evidence proportionate to claims
State your assumptions, reproduction steps and limits. Write the expected result before a test, so failures cannot be renamed.
Accurate attribution
Every contribution is credited by role, with permission. Human and AI contributions are named the same way.
Respect other people's information
Use synthetic data. Never submit records about people who have not agreed to it.
No hidden vetoes
Sponsors and implementers, including EPMA, get no undisclosed say over findings. Licence terms are published before contributions are accepted under them.