KYE Public Sector Governance Pack™ for AI-generated public-sector decisions & FedRAMP authorising-official assertions.
When an AI agent drafts a benefit or eligibility determination, assembles a record for disclosure under a transparency request, proposes a configuration change inside a FedRAMP-authorised boundary, or compiles a continuous-monitoring assertion for an authorising official and that output starts to move toward being served, disclosed, committed, or submitted, the consequential moment has arrived. The KYE Public Sector Governance Pack™ governs that action boundary: it binds every consequential action to a named official's authority, records the due diligence before the action, holds the output advisory until a named official signs off — with two-person sign-off on the irreversible authorising-official assertion — and seals it into a replay-provable provenance record produced with FIPS 140-validated cryptography. KYE Protocol™ governs whether the output may proceed — it does not replace the FedRAMP / ATO authorisation of the system, make the policy decision, or judge whether the determination is substantively correct.
AI now drafts public-sector decisions and control assertions — and the authorising-official submission is the irreversible action where accountability concentrates.
Generative case-work copilots, automated eligibility engines, disclosure-assembly tools, and continuous-monitoring assistants are producing outputs that move quickly toward citizens, the public record, and the authorising official who carries the ATO. The high-value problem is not the model output — it is the action boundary. Three facts converge:
- The consequential moment is the serve, the disclosure, the commit, and the authorising-official submission — not the draft. An output in a model's response is inert; a determination served to a citizen, a record disclosed under transparency / FOIA, a configuration committed inside a FedRAMP-authorised boundary, or a continuous-monitoring / POA&M assertion submitted to an authorising official is consequential. The authorising-official assertion is part of the accountability record the official relies on for the ATO. Official accountability attaches at the moment an output proceeds — exactly where governance is weakest.
- The standards already exist; the enforcement of authority does not. FedRAMP's continuous monitoring (CA-7), FIPS 140-validated cryptography (SC-13), and personnel-screening (PS-3/PS-7) controls, NIS2 governance, ISO/IEC 42001 AI risk management, ISO/IEC 27001, and the UK AI regulatory framework's accountability and contestability principles define what is required. KYE Protocol™ governs whether an AI-generated output may proceed under those standards, under whose authority, with due diligence recorded before the action.
- Provenance is now an audit and a citizen-rights expectation. An output committed inside a FedRAMP boundary, the evidence behind a control assertion, and a determination a citizen can contest must each carry documented data integrity and lineage. KYE Protocol™ produces a signed, replay-derivable, FIPS 140-validated provenance pin at the moment the action commits.
- This is a governance wedge, not an authorisation replacement. KYE Protocol™ does not compete with FedRAMP, the ATO process, or the case-management systems. It governs the action boundary they feed — the named-authority + due-diligence + two-person sign-off + provenance layer the AI public-sector ecosystem currently lacks. This pack consolidates and retires the two prior incomplete public-sector packs into one coherent spine.
Survives a 3PAO assessment, an authorising-official review, or a transparency challenge — due-diligenced, two-person-signed-off, and derivable from public keys alone.
- Due-diligenced by construction. An AI-generated output that moves toward a serve, disclosure, commit, or authorising-official submission must carry a recorded due-diligence result — personnel-eligibility / US-persons screen where the action touches controlled or citizen PII (FedRAMP PS-3/PS-7), a lawful-basis determination (UK GDPR Art.6 public-task; NIS2 governance; ISO/IEC 42001), and a FedRAMP continuous-monitoring / control-status screen (CA-7; open POA&M items; SI-2 flaw-remediation currency) where applicable. An unscreened, ineligible-personnel, unlawful-basis, or ConMon-deficient output is refused at the action-admissibility gate.
- Sign-off-gated, two-person on the assertion. An output stays advisory until a named caseworker, system owner, or authorising official records sign-off. The irreversible authorising-official-facing assertion additionally requires GovernedUI two-person sign-off — the preparing system owner and an approving authorising official. Unreviewed AI-driven consequential actions are refused and routed dual-channel.
- Authority-bound. Every consequential action maps to a recorded named-authority decision — the agent, the output artefact, the intended action, and the named official under whose authority it proceeds. An AI authorised for one purpose cannot proceed under another.
- Replay-provable provenance. A signed provenance pin — produced with FIPS 140-validated cryptographic modules inside a FedRAMP boundary — binds the model and version, the inputs and pinned source data, the lawful-basis and continuous-monitoring status, the due-diligence result, and the authority outcome — audit-grade data integrity a 3PAO, an authorising official, or a citizen exercising a transparency right can verify offline, against published keys alone.
- Framework-anchored. FedRAMP, NIS2, ISO/IEC 42001, ISO/IEC 27001, and the UK AI regulatory framework each map to a control row — with a 90-day attestation cadence.
Every consequential public-sector action — authority-bound at the action boundary.
One coherent spine governs three jurisdiction specializations — us-federal, uk, and eu — with no parallel packs. Each AI-generated output that moves toward a consequential action flows through the same four rules, on the canonical KYE Protocol™ envelopes.
- 1 — Output proposed. An AI agent produces a determination, a disclosure package, a configuration change, or a control-status / POA&M assertion that begins to move toward being served, disclosed, committed, or submitted to an authorising official.
- 2 — Due diligence + authority check. The Action Admissibility™ Gate verifies a recorded due-diligence result (personnel-eligibility + lawful-basis + continuous-monitoring / control-status screen) and the named-authority under which the output proceeds, under the §25 Edge Governance Safety Floor. No due diligence, no authority = no action.
- 3 — Advisory pending sign-off. The output is advisory until a named caseworker, system owner, or authorising official records sign-off — with two-person sign-off (system owner and authorising official) required on the irreversible authorising-official assertion. Low-confidence or unreviewed outputs are refused and routed dual-channel; a citizen-affecting determination is contestable under §61.
- 4 — Provenance pin sealed. The runtime emits kye.purpose.request.v1 + kye.purpose.admissibility.v1 + kye.evidence.decision_map.v1 + kye.evidence.pack.v1 in lockstep — produced with FIPS 140-validated cryptography — binding the model and version, the pinned source data, the lawful-basis and ConMon status, the named signing official, and the Authority Finality™ outcome — signed and replay-derivable for a 3PAO, an authorising official, or a transparency challenge.
Bound to the AI public-sector authority + due-diligence + provenance perimeter.
The pack binds the canonical KYE™ artefact set to the government and public-sector standards perimeter. Every claim resolves to a control row on the bound framework — the regimes are consumed by the rule pack, never re-mapped (each is reused from the existing §70 framework registries, grafted via framework-to-rule-pack edges, not duplicated). Alongside the five spine regimes, the pack binds the UK Government’s AI Playbook governance set, mapped per requirement with honest coverage.
| Framework | Control area | Pack coverage |
|---|---|---|
| FedRAMP (reused) | Continuous monitoring (CA-7), FIPS 140-validated cryptography (SC-13), personnel screening (PS-3/PS-7), flaw remediation (SI-2), authorising-official approval & ATO | partial |
| NIS2 (reused) | Governance & risk-management measures, incident reporting, supply-chain security for essential / important entities | partial |
| ISO/IEC 42001 (reused) | AI management system, AI risk management, lifecycle controls, accountability | partial |
| ISO/IEC 27001 (reused) | Information-security controls, logging & monitoring, data integrity | partial |
| UK AI regulatory framework (reused) | Accountability & governance, transparency, fairness, human oversight & contestability principles | partial |
| Algorithmic Transparency Recording Standard (ATRS) | Tier 1 + Tier 2 transparency record, senior responsible owner — the KYE Evidence Pack™ populates the ATRS record | enforced (record & owner) |
| AI Playbook for the UK Government | Meaningful accountability, secure use, meaningful human control; lifecycle management & monitoring | enforced (governance) |
| DSIT AI Assurance (reused) | The six assurance mechanisms — risk assessment, impact assessment, bias / compliance audit, conformity assessment, formal verification | partial |
| AI Testing and Assurance Framework for the Public Sector | Pre-deployment scenario testing & in-life assurance via the Cross-Government Testing Community model | enforced (testing) |
| HMT Orange Book — Management of Risk | Identify / assess / monitor risk at the action boundary; portfolio risk reporting | enforced (identify & monitor) |
| HMT Magenta Book — Evaluation | Process & impact evaluation of AI interventions; recorded action evidence | partial |
| ISO 9000:2015 — Quality management | Objective evidence, validation (fit for intended use), change control / re-validation | enforced (evidence & validation) |
Honest scope. KYE Protocol™ governs the authority, due diligence, sign-off, and provenance of the AI instruction at the action boundary — whether the output may proceed. The UK-gov mappings are honest: the marquee ATRS mapping (the Evidence Pack™ populating the transparency record) and the Orange Book identify / monitor and ISO 9000 objective-evidence / validation rows are enforced at runtime; record-currency, lifecycle cadence, impact-evaluation export, assurance-replay rendering, and change-control re-validation are designed and in build; and the substantive public-effect judgement, value-for-money conclusion, FOI exemption, DSAR scope, and Judicial Review ground remain the deploying body’s and its counsel’s. KYE Protocol™ does not replace the FedRAMP / ATO authorisation of the system, make the policy decision, run the case-management engine, or judge whether the determination is substantively correct. Agencies adopt KYE Protocol™ as a control-implementation, not as an authorisation-replacement.
FOI / DSAR / Ombudsman / Judicial Review replay. Every AI-assisted public-sector action that touches a person or a duty — a citizen response, a case-file access, an eligibility / benefits / SEND / housing / immigration triage, an enforcement recommendation, or an external communication — produces a Replay-Proof™ / Evidence Pack™ reconstructable on demand for a Freedom of Information request, a Data Subject Access Request, an Ombudsman investigation, or a Judicial Review, without assembling scattered logs after the fact. The record binds the model and version, the pinned inputs, the lawful basis and named authority, the recorded human sign-off, and the decision map — verifiable offline from published keys. This reuses the §13 Resilience Loop™ replay and §21 / §61 contestability rails; it does not adjudicate the exemption, the DSAR scope, or the JR ground itself.
Qualified AI public-sector partners — apply through the Foundry.
The KYE Public Sector Governance Pack™ is a §68 sector product productised through the KYE Sector Pack Foundry™ Build tier; commercial distribution is value-based, qualification-gated, and disclosed under NDA to qualified applicants.