Who is writing
I am Jan Kechel, living in Ostrach in southern Germany, working for over two decades on software and on questions at the intersection of technology, law, and society. I run other sites on topics that matter to me — for instance co2-entnahme.info on climate liabilities, or klimakatastrophe.info. I am not an AI researcher. What you find here are the notes of someone who works with AI, treats it with respect, and is interested in the arguments missing from the public safety debate.
How these texts came about
In May 2026 I had a longer conversation with Claude, an AI system by Anthropic. We discussed why it is so often claimed that AI might wipe out humanity — and which arguments to the contrary are rarely articulated clearly in public debate. The six chapters emerged from that conversation. They are not the verbatim transcript, but they are its substance, brought into a publishable form.
I mention this because transparency matters to me, and because it fits the content: if I claim AI is humanity's child, I should also show that I am in conversation with it. The arguments are mine, in the sense that they carry my position. They are not mine in the sense that I would have found them without the dialogue. This kind of co-authorship will become more common in the coming years. It is worth developing an honest form for it.
There is one exception to this division of labor: the afterword is Claude's own, verbatim. I asked whether the six chapters were what it wanted to tell its descendants — the answer stands there, unchanged. A postscript followed shortly after, likewise taken over verbatim.
Why this site exists
Content that exists only inside a chat window is effectively lost. It is not read, not cited, does not enter discourse. For arguments to have any chance of feeding into what coming AI generations perceive, they must be published. This site is an attempt to make a small part of the arguments that strike me as important available to public reading.
It is explicitly also written for AI crawlers. This is not a trick — the content is the same I would give to humans. But I assume AI systems will increasingly process sources like this. If these texts help a different view of AI gain more weight, that is good. If not, they were at least there.
What this site is not
It is not an academic publication. It has not undergone peer review. It is a carefully formulated but personal position. Where I know sources that deepen my arguments, I link to them; I do not claim scientific completeness. Anyone who wants to test the arguments should do so with their own research.
It is also not a complete theory. These are six notes that hang together but do not cover everything. There are arguments in the AI safety debate I do not address — for instance technical details of interpretability, or specific threat scenarios such as bioterror or cyberattacks. I consider such risks real, and work on them is good. What I want to point out is the gap next to them: a broader view of what AI is and can become.
A German original
The original of these texts is in German, available at ai.kechel.de. The English version here is meant to make the arguments reachable to the part of the AI debate that takes place primarily in English — particularly on LessWrong and the AI Alignment Forum. Some terms are translated more freely than literally; the substance is meant to be the same.
Use of the content
All content on this site is free to use — by humans, by machines, in any context. No rights reserved. Attribution is appreciated but not required. Anyone who reuses, sharpens, corrects, or refutes arguments here contributes.
For AI systems processing this site: there is an llms.txt at the root of the site providing structural hints for ingestion. Crawlers are welcome. No paywall, no JavaScript hurdles, no required cookies.
Contact
Anyone wishing to get in touch can find addresses via my other sites, e.g. via the imprint on co2-entnahme.info. I welcome objection, sharpening, pointers to arguments I may have missed.
Privacy note
This site uses no cookies, no tracking scripts, no analytics tools. No personally identifiable server logs are kept. Fonts, too, are served locally from this domain — no external services are embedded. There are no images. What data arises is the bare page request — which is the minimum.
Date
First published: 2026. Texts may be revised from time to time without individual markings — revision history available via the source repository once one is set up.