AI Usage Policy

I use AI toolsThe tools, as of this revision: Claude (Anthropic), Gemini (Google), and open-weight models such as Kimi (Moonshot AI). Named rather than left abstract, because a list ages honestly. in my work. This page sets out where, and on what terms, so that anyone reading this site knows what they are reading.

The case against these tools is real and I take it seriously. They are confidently wrong at a rate that is difficult to fact-check. Many are trained on other people’s work without permission. The legal position on who owns their output is unsettled. None of that is resolved by enthusiasm.

The case for them, in my hands, is narrower than the hype suggests and more useful than the dismissals allow. They are good at the parts of a job that are laborious rather than interesting: refactoring code, reading configuration files, pointing at sources I would otherwise have spent an afternoon finding.

So this is not a policy of abstention, and it would be dishonest to publish one. It is a policy of disclosure and retained responsibility.

The policy

  1. I am responsible for everything published here. Whatever drafted a sentence or a function first, I have read it, understood it, and tested it. If something on this site is wrong, that is my error and not a tool’s. “The AI wrote it” is not a defence I will offer.
  2. Writing published under my name is mine. Essays, newsletter issues, captions and notes are written by me. I use these tools to argue with, to check a fact, or to be told a paragraph isn’t working. I do not use them to generate the prose. That keeps the ownership of my writing unambiguous, and it is the basis on which you can trust what you read here.
  3. All photographs are mine. Every image on this site, and at the head of every issue of Interoperable, is a photograph I took. No generated imagery appears here.
  4. Code and infrastructure work is often AI-assisted, and logged. A good deal of this site has been built or rebuilt with AI assistance. That work is recorded in the ChangeLog, which is the disclosure. Nothing is committed that I have not reviewed and run.
  5. Research is verified before it is used. These tools are a reasonable way to find a source and a poor way to trust one. Anything I repeat here has been checked to the best of my abilities.
  6. I will not pass off generated work as craft. Not my own, and not anyone else’s. Where a tool has done something substantive, it says so.
  7. I use tools I have reservations about.Not every tool is on the table. I will not use anything from xAI, or from the companies it shares an owner with. The technology is never the whole of the decision: who builds a thing, and how they conduct themselves in society, is part of what you endorse by using it. Others will draw that line in a different place. This is where mine is. The training-data question is not settled, and I have not settled it in my own favour. I have decided that assistance with code and preliminary research is worth that compromise. I will never compromise on writing or photography. I write the words. I take the pictures.

Where this came from

The structure of this document, and a good deal of its reasoning, is adapted from Complexly’s AI Policy, written for a media company producing educational videos at scale. Their conclusion is near-total abstention in production, which is right for their circumstances and not for mine. I have kept their argument and adapted the answer to fit my needs and morals. Credit where credit is due.

Revision one. 1 3 5 7 9 10 8 6 4 2 — . This policy will change as the tools do. Revisions to be dated.