Right to human oversight

What it is

The right to human intervention is the right to request that a natural person, not an algorithm, review, supervise or take a decision when you are subject to artificial intelligence systems, especially if:

• the AI decision has a significant impact on your rights;
• you believe the AI decision is wrong, unfair or discriminatory;
• the situation requires an understanding of context that the AI does not have.

What the AI Act requires (Art. 14)

High-risk AI systems must be designed with effective human oversight, allowing the person to:

• understand the AI’s capabilities and limits;
• remain aware of the automation, without over-relying on it;
• correctly interpret the AI’s outputs;
• decide when not to use or follow the AI;
• intervene in or stop the system;
• override the AI’s decision.

Three decision models

AI-only decision

• The algorithm decides on its own.
• No person is involved.
• The AI’s output is the final decision.

AI decision with human oversight (what the AI Act requires)

• The AI provides a recommendation or proposal.
• A person reviews the output.
• The person has final authority over the decision.
• The person can depart from the AI.

Human decision assisted by AI

• The person uses AI as a support tool.
• They keep full control.
• The AI does not take decisions.

Levels of human oversight

• Human-in-the-loop: every AI decision requires human approval; the person reviews it before it is executed. Maximum control.
• Human-on-the-loop: the AI operates autonomously; the person monitors and intervenes when necessary.
• Human-in-command: the person can switch off or modify the system; final responsibility and strategic control remain human.

The AI Act requires at least one of these levels for high-risk systems.

Where human intervention is mandatory: high-risk AI systems

Employment and HR

AI systems for recruitment

The scenario: you apply for a job; the AI screens CVs, analyses video interviews, assigns a score and “recommends” which candidates to invite.

Your right:

• the final decision cannot rest with the algorithm alone;
• a recruiter must review the AI’s output;
• you can ask: “I want a person to review my application”.

How to exercise it:

“Regarding my application for [position]: if an AI system was used for screening, under Art. 14 of the AI Act I request 1) a human review of my application; 2) an assessment not based solely on the algorithmic output; 3) an interview with a natural person, if foreseen at the next stage. The AI Act requires human oversight of high-risk AI systems, such as those used for recruitment.”

AI systems for performance evaluation

The scenario: the AI monitors employee productivity, assigns a rating and suggests promotions or dismissals.

Your right:

• a manager must review the AI’s assessments;
• you can contest the rating and ask for a human review;
• final decisions (promotion, dismissal) must involve a person.

Credit and finance

AI systems for creditworthiness assessment

The scenario: you apply for a mortgage or loan; the AI analyses your creditworthiness and produces a score and a recommendation (approve or reject).

Your right:

• the final decision cannot be the AI’s alone;
• a bank officer must review it;
• you can ask: “I want a human review of my application”.

How to exercise it:

“Regarding my loan application of [date]: if the negative decision is based on an AI system, under Art. 22 GDPR and Art. 14 of the AI Act I request 1) a human review of my file; 2) an assessment that considers the elements the AI did not capture; 3) a meeting with an officer to explain my situation. I am available to provide additional documentation.”

AI systems for insurance pricing

The scenario: you apply for a policy; the AI calculates a very high personalised premium, or refuses cover.

Your right:

• a person must review the price calculated by the AI;
• you can contest it and ask for an explanation from a person.

Education

AI systems for assessing students

The scenario: an exam is marked by AI, which assigns the grade; the grade decides admission or failure.

Your right:

• a teacher must review the AI’s marking;
• you can ask for a human review if you believe the grade is unfair;
• critical decisions (such as university admission) must involve a person.

How to exercise it:

“Regarding the assessment of my [subject] exam of [date]: if the marking was carried out by an AI system, I request a review by a qualified teacher. Art. 14 of the AI Act requires human oversight of high-risk AI systems in education.”

Public services

AI systems for access to social services

The scenario: you apply for a benefit; the AI checks the requirements, decides eligibility and denies access.

Your right:

• a public official must review the decision;
• you can ask to be heard by a person;
• the final decision must take your particular circumstances into account.

AI systems for allocating public housing

The scenario: you apply for social housing; the AI assigns a priority score and draws up the ranking.

Your right:

• a committee must review the rankings produced by the AI;
• you can contest your position and ask for a new human assessment.

Justice

AI systems supporting judicial decisions

The scenario: the AI assesses a defendant’s risk of reoffending and suggests a sentence or a precautionary measure; the output influences the judge.

Your right, reinforced:

• the judge always decides; the AI is only a support;
• the judge cannot delegate the decision to the AI;
• the defendant has the right to a fair trial, with a human decision.

Healthcare

AI systems for medical diagnosis

The scenario: the AI analyses X-rays and tests, suggests a diagnosis and recommends a treatment.

Your right:

• a doctor must review the AI’s output;
• they cannot rely blindly on the AI;
• the patient has the right to a human second opinion.

When to ask for human intervention

The AI decision seems wrong

Typical reasons:

• the AI did not consider particular circumstances;
• it relied on incomplete or outdated data;
• it applied rigid criteria that do not fit your case.

Example: the bank’s AI rejects your loan because your credit score is low. But you have just changed jobs and your income is now stable: the AI did not consider the new situation. Ask for a human review that takes the context into account.

The AI decision may be discriminatory

Typical reasons:

• the AI may be biased on gender, age or ethnic origin;
• the outcome seems unjustifiably negative;
• other people in a similar situation were treated differently.

Example: an HR AI rejects your application and you suspect gender bias, in a male-dominated sector. Ask for a human review, free from algorithmic bias.

The situation is complex and the AI does not understand it

Typical reasons:

• the context requires empathy and human understanding;
• there are exceptional circumstances the algorithm did not foresee;
• there are nuances the AI does not grasp.

Example: an insurer’s AI calculates a high premium based on age. But you are an ultramarathon runner in excellent health: the AI only considers age, a statistical factor. Ask for a human assessment that takes your actual health into account.

The decision has a very significant impact

• A dismissal.
• A mortgage refused for your home.
• Failing a decisive exam.
• An essential benefit denied.

For decisions with a strong impact on people’s lives, the right to human intervention is reinforced.