The Missing Place for “Stop”: Why systems that work correctly still fail

Key points

• Drift is rarely sudden: decisions begin to move faster than the organisation’s ability to own them.
• AI does not introduce the flaw; it accelerates it by compressing the distance between insight and execution.
• Accountability that appears only after the fact is symbolic rather than operational.
• Every component can be locally correct while the whole system loses coherence.
• The test: is there a structurally empowered place where a decision, including an AI-supported one, can be stopped before execution — and is someone accountable for invoking it?

Decisions without a place to be stopped

Organisations combine formal mechanisms — roles, procedures, committees, approvals — with informal ones such as experience and tacit judgement. The balance breaks when decisions move from analysis to action almost automatically. AI often shortens that path. That is not dangerous in itself; it becomes dangerous when the architecture never defines the moment at which someone is structurally empowered to say “stop”.

Ownership before outcomes

Responsibility is usually assigned once consequences are visible, through audits, reports and lessons learned. The critical moment is earlier: when a decision shifts from possibility to reality. If no person, role or mechanism is responsible for intervening at that point, responsibility is deferred into the future, and AI simply follows the path the organisation implicitly allows.

Local correctness, global incoherence

Models can be statistically sound, processes can follow their steps and teams can act in good faith within their mandates, while the whole system loses coherence. It then becomes hard to answer basic questions: why was this decision made, who had the authority to stop it, and when did responsibility move from human judgement to system execution? If a system cannot answer them, the issue is structural rather than a question of intelligence.

What the paper does not claim

This is a short conceptual essay. It identifies a structural gap and poses the questions an organisation should be able to answer; it does not propose an implementation.

This page is a summary written by One World AI. The paper on Zenodo is the authoritative version.

Cite this paper

Bański, P., & Sobieski, S. (2026). The Missing Place for “Stop”: Why systems that work correctly still fail. Zenodo. https://doi.org/10.5281/zenodo.18188447

More research

• Standing to Stop — A stopping point is not enough. The party at it needs presently valid, traceable authority, and the stop must actually change what can happen.
• Architectural Constraints in AI Governance — Governance of AI-enabled operational systems has to live in the architecture that decides which state transitions are admissible — including across sequences of individually acceptable steps.
• Enforceable Decision Boundaries — Inference may propose actions but never authorise them. Authority is resolved outside the model, and execution is impossible without it.
• Prompt Engineering as a Control Illusion — Prompts influence what a model says. They cannot decide what a system is permitted to do.