AI Attribution Policy
Whether and how I use AI tools is my business, not a public disclosure item. Why attribution enables malicious profiling, and what I'll do about it.
Technical

Introduction

There’s a growing convention that anyone publishing work — code, articles, designs, analysis — should disclose whether AI was used in producing it, and often which AI. Some people attach labels. Some embed it in metadata. Some are asked for it explicitly.

This post is my position on that, and it’s a negative one: I don’t disclose it, and I’m not going to.

It’s my business, not yours

Which tools I use, which models I run queries through, and what I put into them is personal information. It sits in the same category as my medical records, my financial position, or trade secrets I hold in confidence. None of those are “transparency” items. None of them become a public disclosure obligation because other people find them interesting.

Why do they sit in that category? Because they describe me. A medical record describes my body. A trade secret describes how I operate. And my AI toolchain describes my habits, my dependencies, and my vulnerabilities.

Why disclosure is a security problem

Attribution is a profiling mechanism. The moment I attach my name to “I used model X via toolchain Y,” I’ve handed anyone who reads it a set of questions worth answering about me:

  • What do I care enough about to put into these tools?
  • What infrastructure do I depend on, and who operates it?
  • What happens if that operator is compromised?

Consider the concrete case. Suppose the tool or model I publicly attributed my work to is compromised — a vendor breach, a poisoned model supply, an exposed API endpoint, a leak of prompt and response logs. An attacker with that data now has something better than a generic threat model: a confirmed list of people who actually used it, what they asked it, and what it told them.

Anyone who publicly identified themselves as a user of that system is instantly short-listed as a potentially fruitful target. I would go from “a person with a public IP and a web presence” to “a person whose toolchain we know, whose conversations we may have, and whose follow-on attacks we can tailor.” I would receive undue attention I never asked for, and I would have to carry a security burden — additional verification, additional paranoia, additional countermeasures — imposed on me by a disclosure I made for zero benefit.

That’s not an acceptable trade. Publishing attribution buys nothing for me and creates a durable, permanent artefact that outlives the toolchain I actually use, and that I can’t recall.

The counter-arguments

“Transparency builds trust.” Trust is established by the consistency and quality of what I publish, over a long record. Nobody’s trust in my work rests on knowing which IDE or editor I use, and it shouldn’t rest on knowing which model I used either. The tool is not the credential.

“Disclosure protects the author from provenance disputes.” If there’s a legal question about the provenance of a specific piece of work, it’s between the parties involved, with evidence exchanged in the relevant forum. A public label isn’t evidence in either direction.

“Undisclosed AI is a safety concern.” Everything I publish already carries my name and my reputation. I stand behind the quality of what I put out: I review it, I test it, and I own it. That’s the accountability that actually protects readers, and it doesn’t change whether a machine helped me draft it.

What I’ll do

  • I don’t label my work with AI attribution.
  • I don’t publish my AI toolchain as part of my infrastructure write-ups. What I publish about my systems describes what I operate; it doesn’t describe everything I use.
  • I review and own everything I publish. The quality bar is the same whether the first draft was mine or a model’s.

The fact that I use particular tools is a detail of my private working practice, no more a matter for public record than the fact that I use a particular editor, a particular database, or a particular accountant.

It’s my business. Not yours.