Security
Where the line sits, and how to report a vulnerability.
Draft policy, awaiting approval by counsel and Systems Engineering. Figures in square brackets are proposals.
Report a vulnerability
We will acknowledge your report within [three] working days and give you a first assessment within [ten]. We will tell you when it is fixed and, with your consent, thank you publicly.
When researching, please do not access other people’s data, disrupt any service or try to trick our staff. Good-faith research within this policy will not be pursued [wording to be approved by counsel].
The policy
| Part | What it says |
|---|---|
| Contact | The address above, also in our security.txt file at the site root |
| Scope | This website, CES products, CIX and, as they open, every interface in this hub |
| What we commit to | Acknowledge within [three] working days; a first assessment within [ten]; fix and disclose on a stated timetable |
| Safe harbour | Good-faith research within the policy will not be pursued, subject to counsel’s wording |
| What not to do | No access to other people’s data; no service disruption; no social engineering |
| Recognition | A public thank-you list, with the researcher’s consent |
The days in square brackets are proposals for counsel and Systems Engineering.
What we publish, and what we protect
Our seventh design principle draws the line: interfaces, the Invariant and the audit method are published; weights and participant data are protected. The hub adds a third category, for material shared under a written agreement.
| Side | What belongs there | Who sees it | How |
|---|---|---|---|
| Published | Specifications and interfaces; the Continuity Invariant; the audit method; illustrative examples; aggregate figures with no names; status and changelog | Everyone | This hub, under the documentation licence |
| Accredited | Real proofs and their inputs; full ledger read; sandbox environments; preview interfaces | Accredited auditors, the Audit Council, deployment partners | Under a written agreement that sets purpose, duration and confidentiality |
| Protected | Model weights; participant data; identity and verification data; signing keys; data in sovereign deployments; vulnerabilities not yet fixed | No one outside the teams that need them, and the Audit Council where its terms allow | Never published in the hub |
Privacy for developers
Any interface that touches personal data will state, on its page: what personal data it handles; the consent rule that governs it (the Participant Charter, the Contributor Rights Charter or the CES privacy notice); where the data is stored; how long it is kept; and how a person can see, correct or delete it. If an interface handles no personal data, its page will say so.
The AI Constitution for builders
Developers who build on CRIS systems, or to our specifications, inherit responsibilities. Three things to know from our proposed AI Constitution: the fifteen actions no CRIS system may take, quoted in full on the Constitution’s pages; the requirement that a human stays accountable for any automated decision that materially affects a person; and the two gates every change must pass. The Constitution is proposed until our board adopts it.
Verification keys
When our first real proofs are signed, we will publish the public keys that verify them here, with fingerprints, validity dates and rotation history. Private signing keys are never published. If a key were ever compromised, we would announce it here and in the changelog, name the affected period, and publish the replacement key.
