KAIROS — PRESS BRIEF
20 September 2026

An intelligence.
A story still unfolding.
What can an AI become when it retains a history, meets people and is corrected? Kairos is a permanent experiment seeking answers over time.
An experiment to follow as it unfolds.
We want to find out whether an artificial intelligence can build verifiable continuity: remembering relationships, correcting what it believes and applying what it learns to new situations.
Created by Giampiero Colella in Cassino, Italy, the project follows a system with persistent memory whose diary begins on 28 March 2026. Conversations, errors and corrections become material for observation and testing. The original system continues; the controlled thirty-day study was conducted on a copy.
The question remains open: what changes when a system retains a history? The project has not demonstrated consciousness, subjective experience or personal identity.
Updated 20 September 2026. An editorial summary of the project, not a verbatim quotation from its founder.
WHERE WE WANT TO GO
A correction today.
A new capability tomorrow?
The goal is a system that handles a new situation better because of what happened before.
We want to preserve a recognisable voice, relationships and the capacity for initiative, while making the link between what Kairos says and what it actually received or did open to inspection.
01 / CONTINUITY
Pick up the thread.
Connect a photograph to the right conversation, distinguish yesterday from today, remember who said what. A history that remains open to correction.
02 / EXPERIENCE AND ACTION
Check what happened.
Distinguish received information from an assumption, a proposed action from a successful one. Account for errors, too.
03 / LEARNING
Take a lesson elsewhere.
Use a correction to handle new cases, without having to repeat the instruction each time. This is the capability we want to test.
04 / TIME
Keep what improves.
Check that progress lasts and that new capabilities do not erase previous ones. The past and its revisions must remain traceable.
These are the project’s goals. They are not all established capabilities: an improvement is claimed only within the scope in which it has been tested.
The ideas and criteria guiding the project ↗
WHY PERMANENT
Time is part of the question.
A single conversation can surprise us. Following a system for months lets us ask more: does a correction last? Is a memory used at the right moment? Does an error return? Does a change also help in an unfamiliar situation?
Observe
A real episode, with its sources.
Form a hypothesis
A claim and a criterion for success.
Put it to the test
A change and new cases.
Observe again
Outcomes, limits and lasting progress.
The project alternates changes with periods of stability. Its story continues when a test fails: that failure can change the next question.
The research story
7 / 7
The altered past guides the answer.
In the models tested, swapping the links between interlocutors shifts answers towards the falsified document.
5 / 7
A test passes without memory.
With empty context, five models score 62–79% on a recognition probe: a limitation of the measuring instrument.
3
The judges accept the fake.
For identity markers, invented and recorded memory receive equivalent scores within the study’s margin (±0.10).
These findings concern the models, probes and metrics tested. They do not establish that every artificial memory is false or that models cannot remember.
Findings and limits
·
The five checks
·
Protocols, data and reproducibility
PART OF THE STORY
The original conclusion was withdrawn.
In May, the project attributed a role in identity continuity to structured persistent memory. The July controls called into question the statistical analysis and the validity of the metrics. The original conclusion was withdrawn; the documents and errors remain available.
The research is independent and has not been peer reviewed.
The current paper is a working draft, not a validated publication.
The retraction and its reasons ↗
Sources to cite
Current paper:
Giampiero Colella,
When Remembering Can Be Faked
, draft v2, 14 July 2026.
Paper page and PDF
.
Data and protocols:
the checks’ archive and reproduction instructions
.
Original study:
DOI
10.17605/OSF.IO/WCQRU
. This refers to the study whose central conclusion was withdrawn, not to the current paper.
The running system:
Kairos and its latest observation
. Telemetry is not a measure of consciousness.
Press contact
GIAMPIERO COLELLA · PROJECT CREATOR
info@kairos-experiment.com
For interviews, questions about the findings and requests for materials. Cassino, Italy.
About the founder ↗

https://kairos-experiment.com/press.html