Deny by default, scoped at sign-in, sealed after the fact.
Governance in VXCTRL is four things working together: a rule set that starts from no, a session that carries its own boundaries, a gate that decides what becomes of-record, and a record that cannot be quietly rewritten afterwards.
What does deny-by-default actually mean here?
That the starting position is no. An action is not permitted because nothing forbids it; it is permitted only because a rule allows it. Rules are priority-ordered, so conflicts resolve deterministically rather than by whichever rule happened to be evaluated last, and the rules themselves are audited.
Grants are scoped by level and by address class. A grant is therefore not a global switch that follows someone around the enterprise: it carries the boundary of where it applies. Combined with the six-level tree, that means a permission is always a permission somewhere, never everywhere.
No, until a rule says otherwise
Nothing is allowed unless a rule allows it. Absence of a prohibition is not permission.
Priority-ordered rules
Rules carry priority, so the outcome of overlapping rules is determined rather than incidental. The rule set is audited.
Bounded by level and address class
A grant applies where it was scoped to apply, by level in the tree and by address class, not across the whole enterprise.
What does a scoped session carry?
Exactly what that person can see and manage, no more, and it expires and refreshes on a short TTL. The scope is part of the session rather than something re-derived from a cache of unknown age.
The short TTL is what keeps the scope honest. A session that lived for a day would keep describing a position its holder may no longer occupy; one that refreshes frequently reflects the tree as it currently is. Identity supports this with verified sessions, argon2id password hashing, signed session tokens, and sessions that can be revoked.
How does something become of-record?
Through an approval gate, and only through an approval gate. Facts recorded in DIN CTRL, the evidence record of what people and AI agents did, are promoted into the system of record by a gate that checks the approver again, requires cited sources, and grades the trust of the fact from those sources.
The gate re-checks the approver
Being the approver earlier in the process is not sufficient. The approver is checked again at the point of promotion.
Sources must be cited
A fact without cited sources does not pass the gate. The citation is a condition of promotion, not a nice-to-have field.
Trust is graded from those sources
The gate grades trust as receipted, logged or inferred, derived from the sources cited, and a client is never allowed to assert its own trust grade.
Raw dumps are refused
The gate refuses raw data dumps. Promotion is for facts with provenance, not for bulk material pushed at the system of record.
Two-person controlBuilt, not yet enabled
Sensitive actions can be made to require a second, distinct approver, who co-signs the exact same write, bound byte-for-byte, so the write that is co-signed is provably the write that was shown. This is built and is not yet enabled, so it should not be counted as a live control today.
What can you prove afterwards?
What was attempted, what was allowed, what was refused, and by whose authority. Monitoring writes every action and every refusal to a hash-chained operational log; Audit seals the outcome onto an append-only audit ledger.
The two layers do different jobs. The hash-chained operational log makes tampering detectable, because altering an earlier entry breaks the chain that follows it. The append-only audit ledger is where outcomes are sealed: nothing is edited in place, so the history is additive.
Refusals matter as much as approvals. A control plane that only recorded successes would leave the most interesting question, what did someone try that they were not allowed to do, unanswerable.
How are AI actors governed?
As nodes in the same tree, with one addition: AI governance controls which AI models each level may use, and interactions are logged.
Because an AI agent is a NODE, everything already described applies to it without modification: reach from position, deny-by-default policy, scoped session, operational log, sealed ledger. The model-level control sits on top of that, so a level can be permitted to act without being permitted to act through any model it likes.
What is built but not enabled?
Three things, listed here rather than buried, because a governance product that overstated its own controls would be the wrong product to trust.
Bring us the thing you need to be able to prove.
Most governance conversations start from an audit question that was hard to answer. Tell us yours and we will show you what the record would contain.
Questions people ask before they commit
How does VXCTRL policy work?
It is deny-by-default. Nothing is allowed unless a rule allows it, and the rules are priority-ordered and audited. Grants are scoped by level and by address class, so a grant carries the boundary of where it applies rather than being a global switch.
What is a scoped session?
Each sign-in carries exactly what that person can see and manage, and it expires and refreshes on a short TTL. The scope travels with the session rather than being looked up and cached indefinitely, so a change in position takes effect instead of lingering.
Is two-person control available today?
It is built but not yet enabled. When it is turned on, a sensitive action can be made to require a second, distinct approver who co-signs the exact same write, bound byte-for-byte, so the second approver cannot be made to sign something other than what was shown to them.
How does a recorded fact become of-record?
Only through an approval gate. Facts recorded in DIN CTRL, the evidence record, are promoted into the system of record by a gate that re-checks the approver, requires cited sources, and grades trust from those sources as receipted, logged or inferred. The gate refuses raw data dumps, and a client is never allowed to assert its own trust grade.
What are the trust grades?
Receipted, logged and inferred. The grade is derived by the approval gate from the sources cited for the fact. It is not something the submitting client can declare for itself.
What record does a governed action leave?
Two. Monitoring writes every action and every refusal to a hash-chained operational log, and Audit seals the outcome onto an append-only audit ledger. A refusal leaves the same kind of evidence as an approval.
Does VXCTRL govern which AI models can be used?
Yes. AI governance controls which AI models each level may use, and interactions are logged.
Are identity lifecycle and access review available?
They exist as feature-gated modules that are built but not yet enabled, so they should not be counted as live capabilities today.