Every eligibility decision a citizen can appeal — and you can reconstruct
When public-sector AI helps decide who gets a benefit, it is high-risk by classification. KLA holds the determination for caseworker review and seals the lineage an ombudsman, inspector, or appeals tribunal will demand — without forcing the agency into a full technology reset.
01The accountability gap
When AI quietly sets a citizen outcome, the agency owns the consequence
Oversight teams do not need another slide about responsible AI. They need a concrete way to stop, review, and replay the decisions that actually affect a person — before the complaint, not after.
The failure mode that draws scrutiny
An eligibility or casework assistant moves from drafting a recommendation to effectively setting a citizen outcome — and the agency cannot reconstruct what evidence produced it, who reviewed it, or how a person would appeal. In the public sector that is not a bug report. It is an accountability and legitimacy problem.
What due process requires
- Meaningful human oversight on the decision that changes a citizen outcome
- A reconstructable basis: what evidence and context produced the recommendation
- A real appeal path: the person can see, understand, and challenge the decision
Three citizen-facing actions KLA holds before they commit
Eligibility determinations
Recommend, but cannot finalise a benefit decision without a caseworker.
Citizen notices
Draft, but cannot send correspondence or a notice without oversight.
Records disclosure
Cannot release or redact records (FOIA/SAR) without a defensible basis.
02From decision to appeal trail
The four steps that turn an AI decision into a record a citizen can challenge
The EU AI Act lists AI that determines access to essential public services among its high-risk categories (Annex III), triggering obligations a responsible-AI statement does not satisfy: meaningful human oversight, record-keeping, and transparency. KLA turns each into a runtime control point.
- 01 · Classify
High-risk by Annex III
AI used to decide access to essential public services and benefits is high-risk. KLA marks the step so it cannot run the same path as low-stakes drafting.
- 02 · Hold
Caseworker stays in control
The eligibility recommendation, notice, or prioritisation is held for a named reviewer before it commits to a citizen.
- 03 · Seal
Reviewable lineage
Source context, reviewer, and final action are sealed into a record oversight bodies and inspectors can replay.
- 04 · Appeal
A citizen can challenge it
The same sealed record is the basis a person needs to understand the decision and exercise a right to appeal.
sealed · offline-verifiableIllustrative appeal trail, not a customer deployment — it shows how the record is a byproduct of execution, produced before a citizen complains rather than after.
Govern in place, without a rip-and-replace
KLA adds the control path to the public-sector systems you already run: it intercepts the consequential action, routes it to a named caseworker, and seals what an inspection actually asks for — the source evidence, the reviewer log, the disclosure-or-redaction basis, and retention that lets an ombudsman replay the case years later. The obligations of EU AI Act high-risk classification become runtime controls on the decisions that affect citizens, not a policy document filed away.