What is human in command?

Human in command is the principle that a person keeps authority over an AI system, including the choice of whether and how to use it at all. The phrase comes from the European Economic and Social Committee, which called for "a human-in-command approach to AI" in an opinion adopted on 31 May 2017. The European Commission's expert group on AI gave it its working definition in 2019, as one of three ways to arrange human oversight.

Three documents, and the one that is often cited wrongly#

The Committee's own-initiative opinion on artificial intelligence, with Catelijne Muller as rapporteur, was adopted at plenary on 31 May 2017 and published in the Official Journal as C 288 on 31 August 2017. Paragraph 1.6 calls for a human-in-command approach to AI, and paragraph 3.42 repeats the call, stressing that people must be able to keep control of machines. Paragraphs 3.25 and 3.27 attach the phrase to autonomy and control in work. This is the earliest use found in this research.

The expert group's guidelines, published on 8 April 2019, set the phrase beside two others. Human in the loop means a person can intervene in every decision cycle, which the group says is often neither possible nor desirable. Human on the loop means intervention in the design cycle and monitoring of operation. Human in command is the widest of the three, and it includes the decision not to use a system in a particular situation and the ability to override a decision the system has made.

The Committee returned to the idea in its 2021 opinion on the draft AI Act (INT/940, again with Muller as rapporteur), describing its long-standing support for a human-in-command approach and urging that certain decisions remain the prerogative of humans. Its 2025 opinion on pro-worker AI, published as C/2025/1185, is the document this research previously cited as the source. Its adopted text speaks of humans remaining "in control" (paragraphs 1.3 and 2.6). "Human in command" appears in that document only inside an amendment that the Committee rejected, so it is not the source and is no longer cited as one.

Authority is the claim, and competence is what it quietly assumes#

The guidelines define command by what a person is entitled to do: oversee, decide, decline, override. They say nothing about whether the person still can. This research adds that second half, and the addition is interpretation. An entitlement to override means little if the person holding it has lost the practice that would let them tell when the system is wrong. Bainbridge described that problem in 1983, and the deskilling evidence has measured versions of it since. It is also why a human in the loop is not a safeguard by itself, and what meaningful human oversight is meant to add.

A different use of the same words exists on this site. Rule 3 of rules before tools says to put a human in command of each initiative, and means a named, accountable owner. That is an organisational sense, close to accountability. The sense defined above concerns the control a person holds over a particular system.

What the term does not establish#

• No legal force. The guidelines are advisory and the Committee's opinions are consultative. The EU AI Act uses its own vocabulary of human oversight.
• No test of whether it works. Neither source measures whether people in command oversee better than people in the loop.
• Earliest use found, not earliest use. The 2017 opinion is the earliest document checked. An earlier use elsewhere has not been ruled out.
• The guidelines wording is taken from the Commission's web page. The page is undated, and the PDF of the 8 April 2019 text could not be fetched, so the definition has not been checked against the PDF.

Key research and primary sources

• European Economic and Social Committee (2017). Opinion on artificial intelligence: the consequences of artificial intelligence on the (digital) single market, production, consumption, employment and society. OJ C 288, 31.8.2017, pp. 1 to 9.
• High-Level Expert Group on AI (2019). Ethics Guidelines for Trustworthy AI. European Commission, 8 April 2019.
• European Economic and Social Committee (2025). Pro-worker AI, own-initiative opinion on AI and employment and labour market policy (full title in the evidence entry). OJ C/2025/1185.

Related SuperSkills research#

On what oversight has to contain, meaningful human oversight and why the loop alone is not a safeguard. On who can act on a system's output, who can override an AI system and when to override AI. On the person left holding the blame, the moral crumple zone. The wider vocabulary is in the glossary.

About this research#

Rahim Hirji is the author of SuperSkills (Kogan Page, 2026), keynote speaker on AI and human capability, and founder of The SuperSkills Intelligence Company. The 2017 paragraph numbers and the adoption date were read in the Official Journal text on EUR-Lex on 19 October 2026. A source line in the earlier glossary entry, pointing to C/2025/1185, was corrected the same day after the opinion was read and found to contain the phrase only in a rejected amendment. The step from authority to competence is the SuperSkills reading and is labelled as one.

How this research works  ·  Reviewed quarterly  ·  Found an error? Tell me and it is corrected on the page.