Authority-State Attestation via Execution Boundary

Applying railway interlocking principles to AI and digital execution. Presented at the 2026 Korean Society for Railway Spring Conference (paper KSR2026S095).

Event

  • When: 2026 Korean Society for Railway Spring Conference, May 21–23, 2026
  • Where: BEXCO, Busan, Korea
  • Role: Author and presenter
  • Paper: KSR2026S095 — 실행 경계 기반 권한 상태 증명 구조: 철도 인터락 원리의 AI·디지털 실행 적용 (Authority-State Attestation via Execution Boundary: Applying Railway Interlocking Principles to AI and Digital Execution)
  • Author: Yountae Kim (Foresight Oversight / Luralain)

What was submitted

The paper started from railway interlocking. A route does not open because someone once reviewed it. It opens only when route setting, locking, track occupancy and signal aspect hold together at the moment of movement.

The paper asked whether the same discipline could apply to AI agents and automated workflows that trigger payments, deployments, control commands or data egress. Its central question was not whether a request had once been approved, but what can still be proved at the moment of execution.

It proposed an Execution Boundary that re-validates live authority state before an actuator is reached, issues a permit bound to that state, and seals the decision evidence. Public cases (SolarWinds, Knight Capital, OAuth-based delegated access) were mapped to the same gap between approval and execution. It also discussed candidate applications in railway operations.

The paper described itself as an early conceptual study based on design and public cases, without experimental validation in railway operations or large digital platforms.

This summary is kept as it was submitted. Terms used at the time are not updated here.

What was asked

No questions from the floor.

What changed after

  • Scope narrowed. The autumn paper (On Record 02) did not extend the boundary to digital execution in general. It took one decision in rail, whether to keep operating after an AI maintenance prediction, and proposed recheck criteria for it without describing system structure.
  • Vocabulary moved. FO Lens later moved its primary term from execution boundary to execution eligibility (September 2026). The boundary is where execution opens; eligibility is the question asked there.

Limits

  • This record shows what was presented and how it was received in the room. It does not show whether the proposal is correct or useful in operation.
  • The paper was not validated in any operator’s environment.
  • No questions is an absence of response, not evidence of acceptance or rejection.
  • Journal publication status is not recorded here.

Later notes

  • 2026-10-31 — The submitted manuscript spells the author’s English name as “Yuntae Kim”. The author’s name is Yountae Kim. The manuscript itself is kept as submitted.