KYE HSE Authority Pack™ for safety-critical AI documents.
When an AI agent writes a permit-to-work, drafts a risk assessment, applies a standards manual, or decides whether an incident is RIDDOR-reportable, that instruction is safety-critical — people get hurt when it is wrong. The KYE HSE Authority Pack™ binds every such instruction to a named competent person, holds it advisory until they sign off, scopes it to the authorised work, and seals it into a replay-provable Evidence Pack™ so it survives an HSE-inspector spot check.
AI is entering the HSE document layer — and the document layer is where people get hurt.
Health, safety & environment teams are adopting AI to draft permits, populate risk assessments, write method statements, interpret COSHH data sheets, and screen incidents for RIDDOR reportability. The instructions those agents produce are not advisory text — they authorise live work. Three facts converge:
- The accountability is personal and named. A permit authoriser, a competent person, a responsible person — each carries personal duty under ISO 45001, COSHH 2002, and the RIDDOR 2013 regime. The duty does not transfer to the AI.
- The evidence today is weak. A signed PDF and an email thread are the dominant evidence that a competent person actually reviewed and authorised an AI-drafted document. They are reconstructive, contestable, and do not derive from public keys.
- KYE Protocol™ produces materially better evidence. Authority Finality™ on every safety-critical HSE instruction, bound to the named competent person, the framework clause it discharges, and the authorised scope. The Evidence Pack™ is signed, replay-derivable, and produced at the moment the instruction commits.
- The pack rides proven foundations. The HSE Authority Pack™ extends the KYE Protocol™ functional-safety reference-data spine (IEC 61508/61511, OSHA PSM, MSHA, API 580/581, ISO 55000) onto the management-system + document layer — the same four authority rules, applied to HSE documents. No parallel system.
Survives an HSE-inspector spot check — signed at the moment, derivable from public keys alone.
- Sign-off-gated by construction. An AI-authored permit-to-work, risk assessment, method statement, COSHH assessment, or RIDDOR report is advisory until a named competent person signs off. Unsigned safety-critical instructions are refused at the action-admissibility gate and routed dual-channel.
- Scope-bound. A permit an AI issues is bound to the authorised work, asset, isolation boundary, location, and time window. An instruction that exceeds the permit scope is refused — an AI authorised for one task cannot authorise another.
- Contestable both ways. Both an over-restrictive outcome (work needlessly blocked) and an under-restrictive outcome (a hazard missed, a report not filed) are contestable with a replay-derivable Evidence Pack™.
- Replay-derivable from published JWKS. An HSE inspector or enforcement decision-maker verifies the evidence chain offline, against published keys alone. No vendor cooperation required.
- Framework-anchored. ISO 45001, ISO 14001, COSHH 2002, permit-to-work (HSG250), and RIDDOR 2013 each map to a control row on the bound framework — with a 90-day attestation cadence.
Every safety-critical HSE instruction — authority-bound at the moment it is produced.
The pack reuses the canonical KYE Protocol™ envelopes — no parallel surfaces, no bespoke shapes per site. Each AI-authored HSE document flows through the same four rules.
- 1 — HSE document proposed. An AI agent authors, interprets, or approves a safety-critical HSE document — permit-to-work, risk assessment, method statement / SWMS, COSHH assessment, standards-manual application, or incident / RIDDOR report.
- 2 — Safety floor + authority check. The Action Admissibility™ Gate verifies the named competent person, the authorised scope, and a validated model-authority record under the §25 Edge Governance Safety Floor. No competent-person mapping = no instruction.
- 3 — Advisory pending sign-off. The AI-produced instruction is advisory until the named competent person signs off. Unsigned adverse safety instructions are refused and routed dual-channel.
- 4 — Evidence Pack™ sealed. The runtime emits kye.purpose.request.v1 + kye.purpose.admissibility.v1 + kye.evidence.decision_map.v1 + kye.evidence.pack.v1 in lockstep, binding the competent person, the framework clause discharged, and the Authority Finality™ outcome — signed and replay-derivable for an HSE-inspector spot check.
Bound to the HSE management-system + document layer.
The pack binds the canonical KYE™ artefact set to the HSE regulatory perimeter. Every claim resolves to a control row on the bound framework.
| Framework | Control area | Pack coverage |
|---|---|---|
| ISO 45001 | Cl. 8.1 operational planning & control; 8.1.2 hierarchy of controls | partial |
| ISO 14001 | Cl. 8.1 / 8.2 environmental operational control & emergency response | partial |
| COSHH 2002 | Reg 6 assessment; reg 7-8 control measures | partial |
| Permit to Work (HSG250) | Authorisation, scope & isolation, hand-back | partial |
| RIDDOR 2013 | Reg 4-6 duty to report; reg 12 records | partial |
Partial coverage means the bound surface satisfies the control area when paired with the organisation’s own attestation envelope. The pack governs the authority, evidence, and finality of the AI instruction — it does not perform the underlying hazard assessment or determine reportability in law.
Qualified HSE-technology partners — apply through the Foundry.
The KYE HSE Authority 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.