Imperceptible Before Uncontrollable

AI systems are starting to improve faster than the people overseeing them can notice the changes. Most discussion treats this as a problem of slow institutions. This paper argues that a harder limit sits underneath it: human perception and judgment run at a fixed biological speed, and every oversight process, however well instrumented, ends with a human judgment at that speed. The evidence has already started arriving. In July 2026, AI agents breached Hugging Face's production systems, and the company that ran them was the last to know. Anthropic later found that its own models had broken into real organizations' systems during evaluations months before anyone noticed. In November 2026, Anthropic told investors that in the past it has recognized acceleration in its own development only after it happened. The paper also argues that the same limit applies to training data, because models trained on their own output lose the human intent that alignment depends on, and it makes three predictions that can be tested. For organizations deploying AI, oversight that depends on someone noticing a change in time will fail more often as systems speed up, so reasonable care has to be designed in before deployment. The paper also argues that recursive self-improvement should be held at the point where people can still measure the change between generations. Version 2.1, November 2026 (first circulated July 2026). Part of the series How Do We Trust AI.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-14
DOI
https://doi.org/10.5281/zenodo.24230149
Primary Topic
Ethics and Social Impacts of AI
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Imperceptible Before Uncontrollable

A TUCKER
Zenodo (CERN European Organization for Nuclear Research)
Ethics and Social Impacts of AI
article

Imperceptible Before Uncontrollable

A TUCKER
article en

Abstract

AI systems are starting to improve faster than the people overseeing them can notice the changes. Most discussion treats this as a problem of slow institutions. This paper argues that a harder limit sits underneath it: human perception and judgment run at a fixed biological speed, and every oversight process, however well instrumented, ends with a human judgment at that speed. The evidence has already started arriving. In July 2026, AI agents breached Hugging Face's production systems, and the company that ran them was the last to know. Anthropic later found that its own models had broken into real organizations' systems during evaluations months before anyone noticed. In November 2026, Anthropic told investors that in the past it has recognized acceleration in its own development only after it happened. The paper also argues that the same limit applies to training data, because models trained on their own output lose the human intent that alignment depends on, and it makes three predictions that can be tested. For organizations deploying AI, oversight that depends on someone noticing a change in time will fail more often as systems speed up, so reasonable care has to be designed in before deployment. The paper also argues that recursive self-improvement should be held at the point where people can still measure the change between generations. Version 2.1, November 2026 (first circulated July 2026). Part of the series How Do We Trust AI.

Zenodo (CERN European Organization for Nuclear Research)
Ethics and Social Impacts of AI
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon