Moral crumple zone: как разработчики стали буфером ответственности для AI-систем

The term "human-in-the-loop" is commonly considered a safety guarantee: a human in the loop means AI is under control. But data from January 2026 and the concept of "moral crumple zone" show that everything works in precisely the opposite way.

The Moral Crumple Zone

Every modern car has a crumple zone — a specially weakened component that absorbs impact in a collision and protects the driver. In 2019, anthropologist Madeleine Clare Elish from Data & Society took this metaphor and turned it inside out. In complex automated systems, a human often becomes such a zone — except here it is the human who deforms, protecting the system. When an AI system makes a mistake, the human "supervisor" bears the legal, professional, and moral responsibility. The algorithm continues to work; the human bears the consequences.

"Moral crumple zone" — a term introduced by

Madeleine Clare Elish in 2019: a human in an automated system can become an absorbing component that protects not itself, but the system.

Six years ago, this was the hypothesis of one anthropologist. Today, based on fresh data, it is a diagnosis.

Numbers That Don't Reassure

In January 2026, Sonar surveyed more than 1,100 developers about their work with AI assistants. The picture turned out to be mixed:

• 96% are unsure about the functional correctness of code written by AI
• less than half verify AI code before every commit
• most continue to use AI to write core code
• the gap between "I understand I need to verify" and "I actually verify" continues to grow

This is not laziness or irresponsibility — it is a systemic trap. AI accelerates the pace, deadlines shrink, and full code review of code you didn't write from scratch requires time and mental resources that simply don't exist. One of the participants in the Baltes et al. study (March 2026) stated without diplomatic caveats: "They literally use you to critically evaluate their AI slop and give it the next prompt".

Verification Debt According to Vogels

AWS CTO Werner Vogels introduced the concept of verification debt — the debt of verification. AI tools accelerate code writing, but simultaneously slow down understanding: generating became easier than figuring out what exactly was generated. Classic technical debt accumulates in code — it is visible, measurable, you can get rid of it through refactoring.

Verification debt accumulates in heads: unverified assumptions, unprocessed patterns, growing dependence on AI as the sole source of truth about what lies in the repository. According to the same Sonar study, developers increasingly describe a feeling that they understand their own code worse than before. At the same time, responsibility for it hasn't gone anywhere — it simply ends up on the side of the human at the moment the system fails.

What This Means

Human-in-the-loop is not a control mechanism, but a responsibility distribution mechanism. If a human lacks the tools for full verification of AI output and doesn't allocate time for it, they become not a supervisor, but a buffer. A moral crumple zone. And when AI makes a mistake — it is the human who breaks.