solutions · government

Wytness for government

Sovereign-region audit trails. Customer-held keys. Storage in your subscription.

YOUR RUNTIMEWYTNESS · OPERATORYour AI agentsees citizen identifierin clearWytness PII layermint pseudonym · sign event<PII:CITIZEN_*>keys never leave yousignedOperator surfaceholds pseudonym onlywytness_envelope_sigagt_merkle_entry_hashResponsible officerapproves where policyrequires itDTA, NIST AI RMF, CDDO, EU AI Act, ombudsman exportsigned export · pseudonyms only · tamper-evident

the question every regulator asks

Can you produce the evidence?

The DTA's AI Assurance Framework, NIST AI RMF, the UK CDDO Playbook, and the EU AI Act (most public-sector uses are Annex III high-risk) all assume a credible audit trail exists for AI-assisted decisions affecting citizens. Sovereignty cuts across all of it: the audit trail is government data the moment it's written, and most SaaS observability ships prompts and outputs to a vendor tenancy in a foreign jurisdiction. That doesn't work for a process touching Medicare numbers, visa records, or anything Defence-adjacent.

AU AI Assurance FrameworkNIST AI RMF 1.0UK CDDO AI PlaybookEU AI Act Annex III §5(a)EU AI Act Articles 12 + 14IRAP at PROTECTEDFedRAMP Moderate / High

a scenario

An ombudsman inquiry, with and without

A state service-delivery agency triages twenty thousand citizen enquiries a week with an AI agent — extracting identifiers, routing to program areas, drafting responses for officer review. Hosting is Azure Australia Central; the model is a third-party endpoint. The ombudsman asks: for six months, show every enquiry auto-resolved for a class of citizens, prove no citizen identifier left Australia readable, and show which officer supervised each auto-resolution against the policy version in force.

Without Wytness that's a service-bus log (no agents), an Azure OpenAI record (no officers), and a case-management table (no signatures) — with citizen identifiers in clear across all three. With it, one signed export: citizen tokens, agent and officer IDs, tool calls, policy fingerprints, chain links — and the raw events never left the agency's own storage account, because BYOS pointed Wytness at an agency-owned container. The inquiry closes against the signed record.

event.json
Government audit record

// Event identity:

id:"evt_01HX9FBC8R5Q4T9X2P6L7N"
time:"2026-05-22T22:47:12.043Z"
agent_id:"agt_citizen_triage_v1"
source:"agt://citizen-services-agent/v1"

// What the agent did:

tool_name:"document.classify_fetch"
decision:"allow"
source_format:"wytness_agt_py"Python wrapper
responsible_officer_id:"ofr_c8f2"

// Parameters (PII tokenised before this row was sealed):

data.citizen_id:"<PII:CITIZEN_a3f9b1c4>"pseudonymised
data.case_id:"<PII:CASE_TOKEN_8e2d>"pseudonymised
data.classification:"secret"
data.sovereign_region:"azure.au-central"
data.policy_version:"ai-assurance.2026.Q2"

// Integrity proof (3-layer):

agt_merkle_prev_hash:"sha256:f3c7a91b..."chain link
agt_merkle_entry_hash:"sha256:b8f3c7a9..."SHA-256
wytness_envelope_sig:"MEUCIQD8R5Q4..."Ed25519
wytness_envelope_key_id:"key_e92d"

one classified fetch from the scenario — citizen identity arrives as tokens, storage stays sovereign

the deliverable

What you can prove

  • Which agent took which action affecting which citizen, to the granularity the AU AI Assurance Framework and NIST AI RMF Measure category expect.
  • That citizen personal information never left your sovereign environment in clear: only pseudonymised tokens and non-identifying metadata reached Wytness.
  • That every decision affecting a benefit, eligibility, classification, or referral was reviewable by a human officer before it took effect, where policy requires it.
  • That the audit-trail storage sits in your own Azure subscription via BYOS — a Blob container in the region you nominate, Australia Central included — with your own immutability policy if you enable it. Azure Blob is the BYOS provider that ships today; an AWS S3 adapter is on the roadmap.
  • That a parliamentary question, ombudsman inquiry, or FOI request can be answered against a tamper-evident, cryptographically signed record — not logs a vendor could later edit.

how the surfaces map

Three surfaces, one product.

/product/ledger

Ledger

Per-event signing with a key generated and held in the agency's environment; citizen identifiers pseudonymised before egress. BYOS points the raw log at a storage account in the agency's own subscription, in a region the agency nominates. Evidence Packs map to the AU AI Assurance Framework, NIST AI RMF, and EU AI Act Articles 12 and 14.

/product/registry

Registry

Portfolio-wide agent inventory with high-risk classification, tool lists, and model-version history. Anomaly rules flag out-of-policy tool use, off-hours decisions, cross-program retrieval, and behaviour shifts that need the AI assurance committee to re-review.

/product/shield

Shield

Approval gates on decisions affecting citizens — adverse determinations, escalations, inter-agency releases — routed to the case-management workflow the agency already runs, not a new tool procurement has to authorise.

for your auditor

Frameworks that matter here

SOC 2 Type II

Trust services criteria the agency's IT security team already evaluates.

ISO 27001:2022

Referenced across most government procurement.

EU AI Act

Benefits eligibility, migration, justice, education are Annex III high-risk.

GDPR

Article 6(1)(e) public-task basis; Article 35 DPIA duty with AI in the loop.

AU AI Assurance Framework

Wytness evidences the audit-trail and model-risk expectations.

NIST AI RMF 1.0

MEASURE and MANAGE support via the signed trail; GOVERN and MAP stay yours.

the boundary

What Wytness does not do

  • We do not yet hold IRAP assessment or FedRAMP authorisation. The architecture is designed to be authorisable; the certifications themselves are work we will do with the first government customer that needs them.
  • We are not a case management system. Wytness records what your agents did against your case-management or eligibility platform; it does not replace the system of record.
  • We do not make the public-administration judgement on whether a particular AI use is appropriate under your portfolio's law. That sits with your agency's chief data officer, legal services, and the relevant minister.

Questions about a sovereign deployment? Ask us.

We set no cookies. Sign-in and preferences use essential first-party browser storage only — no tracking, advertising, or third-party analytics. Privacy Policy