rotascale
Sign in See it live

Solutions — sector

Airgapped is a control, not a deployment note.

The sector that asks the outbound question first and does not accept a policy statement as the answer. Airgapped mode refuses to start if any setting selects a service outside your network — checked at startup, not at first use.

EU AI Act · national security frameworks National authorities clause map
Status
Varies by jurisdiction and classification
Applies to
Citizen services, benefits, defence and intelligence

What airgapped actually forbids

Every outbound path, enumerated
The connections a deployment could make are listed and probed rather than described. A module that opens a network connection without appearing in that inventory fails the build.
Refusal at startup
With airgapped mode set, the deployment refuses to start if any setting selects a service outside your network. Configuration that appears to take effect and does not is the most dangerous state a control can be in.
And what you give up, stated
External anchoring is an outbound path, so airgapped mode forbids it — which means integrity rests on trusting the operator. The platform says that rather than quietly dropping the property and leaving the claim standing.

No model is called on any code path, airgapped or otherwise. That is asserted by a test which fails the build if a model provider appears in the dependency graph, and it is a stronger claim than having an AI feature.

Equity, and the vocabulary problem

Which attributes may lawfully be used for bias testing is set by the jurisdiction, not by us, and the categories differ between them — caste is a recognised ground of discrimination in some jurisdictions and is neither protected nor lawfully collectable in the EU. A profile carries the attributes, the legal basis for each and the conditions attached, and the platform distinguishes conditions it enforces from ones it can only record.

The agents your teams are about to ship

Public bodies carry an obligation private firms do not: a decision affecting a citizen has to be explicable to that citizen, and to an ombudsman, years later.

  1. 1 authority Is there any authority for this?
  2. 2 status Is the grant active?
  3. 3 window Is now inside the grant's window?
  4. 4 scope Does the grant cover this action?
  5. 5 clean_context Was the context clean, where that is required?
  6. 6 bounds Do the per-action limits hold?
  7. 7 policy Does the policy on the grant permit it?
  8. 8 budget Is there room under the ceiling?
  9. 9 review Does this need a person?
Nine gates, evaluated in this order before the action. The lit ones are the gates that hold the four agents below — which is what differs between one industry and the next. The order does not.
Benefits eligibility agent authority

Assesses entitlement against the rules and issues the determination, at a volume no caseworker team can match.

Consequential action
Refuses or reduces a citizen's entitlement
What goes wrong
Automated decisions about essential services are Annex III high-risk, and an unexplainable refusal is the failure mode that ends up in a public inquiry.
What is recorded
The named official whose authority the agent acted under, the rules in force at the time, and what the citizen was told.
Procurement screening agent policy

Screens bids for eligibility, conflicts and sanctions exposure before a panel sees them.

Consequential action
Excludes a supplier from a competition
What goes wrong
Exclusion is challengeable. A screen nobody can reproduce six months later is an exclusion that will not survive review.
What is recorded
The policy version evaluated, the inputs, and the decision — reproducible rather than remembered.
Case backlog agent window

Works a queue of applications end to end during a surge, clearing what is unambiguous.

Consequential action
Closes a citizen's case without human review
What goes wrong
Emergency authority granted for a surge that quietly persists into normal operations, because nobody set it to end.
What is recorded
The window the authority was valid for, and every call refused outside it.
FOI and disclosure agent clean_context

Assembles responses to information requests and proposes redactions.

Consequential action
Releases a document externally
What goes wrong
A single missed redaction is unrecoverable, and the material that steers the agent may be the requester's own words.
What is recorded
What was retrieved, what was untrusted, and what was released.

Ambition is the point of these: none of them is a chatbot. Each is an agent taking an action with a consequence somebody has to answer for — which is exactly the moment a bounded authority stops being paperwork and starts being the reason the project is allowed to ship.

What an ombudsman or a committee will ask

Not hypothetical questions. These are the ones that arrive in writing, with a deadline, and the honest answer to most of them is a query rather than a project.

Why was this citizen refused?
The trajectory holds the decision, the authority behind it and the rules in force at that moment — not the rules in force today, which is the difference between an answer and a reconstruction.
Who authorised the system to decide this at all?
A grant is signed for by a named accountable human, with an expiry. There is no ambient permission to act.
Can this run inside our network?
Single-tenant in your environment, including airgapped — and the deployment tells you which evidence property you gave up by disconnecting it, rather than pretending nothing changed.
Is the evidence tamper-evident?
Records are sequenced and Merkle-sealed, with inclusion proofs a reviewer can check without access to your systems. A gap in the sequence is detectable; it is not a matter of trusting the log.

Identifiers this deployment verifies here

Every one of these is a checksum, not a shape. The claim is “this is a valid national identifier and the check digits agree”, not “this looks like one” — which is why the count on a screen can be cited rather than caveated.

FamilyHow it is verified
National identifier (PL, BE, NO) — scheme-specific weighted checksum
US Social Security Number — structural rules only — asserted, not observed
Singapore NRIC/FIN — weighted sum, prefix-dependent check letter
Singapore UEN — entity-type check letter
Australian Business Number — modulus 89 after the documented subtraction
Australian Tax File Number — ATO weighted modulus
EU VAT number — per-state check digit where the state publishes one

And what it does not look for. Names, addresses and dates of birth in free text carry no checksum, and finding them needs a model this deployment does not run. Neither are card numbers, medical record numbers, or national schemes beyond the three above. A clean screen means no verified identifier was found — never that the payload holds no personal data.

Written for this sector

14 posts touch this sector — the five most recent are above.

See it on your own problem

Bring the decision you would least like to defend to an ombudsman. We will show you the trajectory it produces — the authority, the rules in force at the time, and what the citizen was told.

A walkthrough is a working deployment with your policy and assurance teams in the room, not a slide deck — bring the agent you are most nervous about and we will put it behind a grant while you watch. If you would rather look first, the demo needs a work address and one click.

Public sector walkthrough See the live demoOr talk to sales →