Human in the loop is not enough on its own.

Human in the loop means a person must approve an AI system's action before it takes effect. Human on the loop means the system acts while a person monitors it and can intervene. Neither is enough by itself. In Eric Anthony Jones's REMAIN CAPABLE, oversight is real only when the person can see, judge and stop the action.

These are working definitions in general use. Human Rights Watch's 2012 report "Losing Humanity" set out a widely cited version of the three categories for autonomous weapons, noting that robotic weapons were already "often divided" this way. The table adapts that framing to AI systems in general.

Being in the loop describes where a person sits in a process. It does not show whether that person can actually judge what they are approving. REMAIN CAPABLE Issue 001 puts it this way:

"Human oversight is not meaningful merely because a person is technically 'in the loop.' I must still be able to see what happened, understand enough to judge it, and have the power to stop, correct, or reject it. Otherwise the final vote can become ceremonial."

— Eric Anthony Jones

Research on automation explains why. Parasuraman and Manzey (2010) reviewed evidence that people who supervise automated systems tend to over-trust them ("complacency") and to follow automated cues without enough checking ("automation bias"). This happens to experts as well as novices, and training alone does not remove it.

European law now names the same risk. Article 14 of the EU AI Act (Regulation (EU) 2024/1689) requires that people overseeing high-risk AI systems are enabled "to remain aware of the possible tendency of automatically relying or over-relying on the output produced by a high-risk AI system (automation bias)", and "to decide, in any particular situation, not to use the high-risk AI system or to otherwise disregard, override or reverse the output".

Issue 001 sets out three conditions. The person needs "enough information and time to judge the output, enough understanding to recognize what matters, and genuine authority to pause, correct, reject, or reverse the action." Without those, Issue 001 says, "'human in the loop' becomes a procedural label rather than a working safeguard."

A confident answer is not evidence. Issue 001 lists the capacity as "Treating a fluent answer as an output to assess, not an authority to obey", and adds:

"The question is not whether to trust systems or distrust them completely. The question is whether trust is calibrated to the evidence and the stakes."

— Eric Anthony Jones

Issue 001 tells the full story behind these lines: a language-learning app moved ahead on its schedule while the learning had not happened.

The system acts with no human input or review in the moment. People may still design, test and audit it, but nobody approves or monitors individual actions.

REMAIN CAPABLE Issue 001 argues the person also needs enough information and time to judge the output, enough understanding to recognize what matters, and genuine authority to pause, correct, reject or reverse the action.

For high-risk AI systems in the EU, Article 14 of the AI Act requires effective human oversight, including the ability to disregard, override or reverse the output and to stop the system. It does not mandate one fixed loop design.

Treat a fluent answer as an output to assess, not an authority to obey. Calibrate trust to the evidence and the stakes, and check what the system actually did before relying on it.

REMAIN CAPABLE — Weekly Intelligence examines what keeps human authority real as AI does more of the work. Issue 001 is free to read before you decide whether to become a Founding Reader ($5 a month).