Human Frontier FR
Menu

Position

AI will never be held accountable. You will.

Knowing where your own judgment stops with AI.

The risk we work on does not depend on an AI error. It is there as soon as someone signs without being able to say what the signature rests on. Researchers call it automation bias: the tendency to follow an automated suggestion instead of checking for yourself. Here, it carries a human signature.

When a decision prepared with AI is challenged, nobody asks the AI assistant to account for it. They ask the person who signed, whether the answer was right or not. That holds for a recommendation memo as much as for an everyday message, or for an answer you keep for yourself and build on later.

The answer does not depend on a moratorium or a breakthrough. Education, so that everyone can see the moment they hand their judgment over to AI. Training, so that they learn to challenge what they read. Strict governance by protocol, so that human oversight leaves a record.

Human Frontier: the discipline of judgment for people who decide with AI.

The Human Frontier model

Two frontiers that delimit three zones

Between the limit of what you know and the limit of what you can still understand, explain, and take responsibility for lies the Dynamic Mastery Zone, where AI helps you explore beyond what you know. The limit that matters is the second one, the Human Frontier: not a fixed line but a state of understanding, personal and shifting. Beyond it, you delegate your responsibility to AI without realising it.

Two frontiers that delimit three zones

Conservation Zone: “Below the knowledge boundary. Territory where I deliberately stay within what I already know. I do not use, or underuse, the potential of AI to explore beyond.”

the limit of what I know

First frontier: the knowledge boundary. “The limit of what I know, the frontier of my knowledge. Some call it the ‘knowledge frontier’.” Beyond it lies the Dynamic Mastery Zone.

Dynamic Mastery Zone: “Between the limits of knowledge and the Human Frontier. This is the optimal zone.” “Territory where AI helps me explore beyond what I know, while maintaining my ability to understand, explain and take responsibility.”

Human Frontier

Second frontier: the Human Frontier. “The limit of what I can still understand, explain and take responsibility for.” “It is set by my role, my context, my judgment.” A state, not a fixed line, hence its blurred edge. Most of the time, nothing signals that you have crossed it.

Delegation Zone: “Beyond the Human Frontier. Territory where I delegate my responsibility to AI without realising it. I validate things I don’t really understand. Control is illusory.”

It is common to work in the Delegation Zone while believing you are in the Dynamic Mastery Zone.

The zone is judged by what you have understood of the document you sign, at the moment you sign it; the profile reads on two dimensions, the reach of your exploration with AI and your awareness of your Human Frontier. The diagnostic, a short questionnaire, reads these two dimensions in what you declare and locates your underlying tendency: the profile that comes back most often, from one subject and one moment to another. A single decision only gives a profile of the moment.

Four ways of working with AI

A map of behaviours, not a classification of people

Everyone moves from one to another depending on the task, the subject and the moment; the questionnaire reads a declared tendency, at a given moment.

Research describes the two excesses: relying too much on automation (Parasuraman and Manzey, 2010) and turning away from it after seeing it make a mistake (Dietvorst, Simmons and Massey, 2015).

REACH OF EXPLORATION WITH AI

extended

limited, close to what I know

Extended reach · low awareness

The Unconscious Delegator

Extended reach · high awareness

The Conscious Navigator

Limited reach · low awareness

The Conservative User

Limited reach · high awareness

The Cautious Retreater

low

high

AWARENESS OF MY HUMAN FRONTIER

Extended reach · low awareness

The Unconscious Delegator

“Explores with AI. Crosses the Dynamic Mastery Zone without pausing. Slips into the Delegation Zone without realising it. Continues to validate without really understanding.” Their confidence does not rest on any verification: as long as nothing goes wrong, everything seems to work. It is when someone challenges the decision that it turns out they could not answer for it. On what they validate without understanding it, what they pass on is correct only when the AI assistant is.

Tell-tale sign “The AI recommended it, so it must be right.”

Extended reach · high awareness

The Conscious Navigator

“Explores with AI. Deliberately leaves the Conservation Zone. Operates actively in the Dynamic Mastery Zone. Detects when approaching the Human Frontier.” When they approach their frontier, they sense it and pause, then choose between Defensive Retreat and Offensive Growth. Their pause is not necessarily long: it can take from a few minutes to several hours, most often within the conversation itself. The time it takes depends mainly on the complexity of the subject: a short pause is no guarantee of quality, and a long pause is no sign of a lack of knowledge. They are not immune for all that: on an evening of fatigue and pressure, they can become an Unconscious Delegator.

Tell-tale sign “Can I learn what I am missing to really understand this and take responsibility for it?”

Their tool for that pause: Riftveil

Limited reach · low awareness

The Conservative User

“Remains systematically in the Conservation Zone. Refuses to explore with AI. Rejects recommendations that go against established practices.” It is not their caution that is the problem; it is the systematic refusal to explore. What they cost shows in what was never tried.

Tell-tale sign “We have always done it this way.”

Limited reach · high awareness

The Cautious Retreater

“Explores with AI. Enters the Dynamic Mastery Zone. But as soon as things get complex, as soon as they approach their limit, they systematically retreat.” They retreat for fear of not being able to take responsibility, and that fear is lucid: they do not sign what they do not understand. Their cost lies in repetition: because they retreat every time, they never take the time to learn what they are missing, and their Dynamic Mastery Zone stays narrow.

Tell-tale sign as soon as the material gets complicated, they bring the recommendation back to what they can take responsibility for with what they already know.

Take the 3-minute diagnostic Twelve statements about your recent decisions · your answers are neither sent nor stored

The pause

The test criterion

The test criterion is simple: can I explain and defend this decision without referring to “AI said”?

The pause: the moment you realise you are approaching your limit, or that you have crossed it. A useful pause fits in one line you can say to yourself: the point at stake, what you are missing, what you choose.

There is another way to miss this moment: sensing that you can no longer follow, and signing anyway. That is no longer delegation: it is putting yourself at risk, and those who, after you, rely on that signature.

The pause, two strategies

Defensive Retreat

“I step back. I simplify. I reduce the scope to bring the decision back to what I can take responsibility for with my current knowledge. I am the one stepping back. The Dynamic Mastery Zone remains what it was.”

Offensive Growth

“I don’t step back immediately. I pause. I ask myself: ‘Can I learn what I am missing to really understand this and take responsibility for it?’” “If the answer is yes, I take the time. I learn. I challenge. I check. I rebuild until I genuinely master it.” “[…] it is no longer me stepping back. It is my Dynamic Mastery Zone that grows.”

From the model to practice: Riftveil

Approaching their frontier, the Conscious Navigator pauses. Riftveil gives them the questions to settle before signing.

A safeguard for anyone drifting towards the Delegation Zone without realising it, Riftveil is also the tool of the Conscious Navigator. Before anyone signs, it examines the AI’s answer on five levels and writes out the questions to settle. It approves nothing.

  1. 1

    Question

    what was actually asked

  2. 2

    Answer

    what the project says

  3. 3

    Reasoning

    how it gets there

  4. 4

    Perspective

    what it did not see

  5. 5

    Applicability

    what it looks like in the real world

“No verdict. The decision is yours.”

Try Riftveil on a project (riftveil.ai)

Why a method is needed

AI models do not reliably check their own work, and people rarely have a method for checking it for them: Huang et al., Google DeepMind, ICLR 2024 · Kamoi et al., TACL 2024 · Tsui, 2025 (average 64.5% blind spot rate, 14 open-source non-reasoning models) · Zhou et al., 2024 · Passi and Vorvoreanu, 2022, overreliance literature · Advait Sarkar, CACM 67(10), 2024 · Randazzo et al., HBS Working Paper 26-021, 2026 (HBR, March 2026).

Human oversight in practice

Article 14(4)(b) of the EU Artificial Intelligence Act provides that the people assigned to the human oversight of a high-risk AI system be able to remain aware of automation bias. Riftveil turns that awareness into a practice: the questions before the signature. And for anyone who will have to answer for their decisions, AttestScope turns it into a signed and dated record that any auditor can verify.

Keeping your judgment in the face of AI is a skill: it can be learned, and it serves at work as much as in private life.

Stay in the loop

New writing, Riftveil releases, sector editions.

Your address is kept by our email service, EmailOctopus, and used for nothing else. Unsubscribe with one click. See the privacy note.

What comes next The diagnostic