Skip to content
Isometric scanner unit: sheets passing through a sealed gate, three stamped copies on the far side

For AI labs

Where your data goes when you hire us

The security questions AI labs ask before sharing a task, answered in full. NDA before anything is shared, work inside your tools when you prefer, no AI tools on your tasks without approval, and your programme never named in public.

Prashant Kumar Singh · · 3 min read

In short

  • An NDA is signed before any task, rubric or output is shared; a sanitised sample is fine if you would rather not share the real thing yet.
  • Work happens inside your platform and repositories when you prefer, so nothing leaves your environment.
  • Clients and programmes are never named in public, no AI tools touch your tasks without written approval, and one person is accountable for all of it.

Before an AI lab shares a task with an outside author, someone in that lab has to answer a security question. Where will the task go? Who will see it? Will it end up in a portfolio? Will it be pasted into a chatbot? This post answers those questions for working with me, in enough detail to forward to whoever has to approve it.

1. NDA first

A non-disclosure agreement is signed before I see any task, rubric, test suite or output. Your paper or mine. If legal review takes time, send a sanitised sample in the meantime: a task with the domain-specific parts replaced. The free 48-hour review works on either.

2. Your platform, when you prefer

Work can happen entirely inside your tools: your task platform, your repositories, your containers, your review queue. Nothing needs to be exported to my machine for the work to be done. If you would rather I work locally, the tasks live in an encrypted volume and are deleted when the programme ends, with written confirmation.

3. Never named in public

Clients and programmes do not appear in my case studies, posts, profiles or proposals. The numbers I publish (acceptance rate, quality rating, tasks accepted on public benchmarks) are aggregate and carry no client name. If a prospective client asks for a reference, I ask you first, and no is a complete answer.

4. No AI tools on your tasks without approval

Some programmes forbid AI assistance on task authoring; some require it; most are somewhere in between. Whatever your policy is, it is followed, and if you have not stated one the default is none. This matters more than it used to: contractors on major training programmes have been dismissed for using AI to produce work that was supposed to be human. The policy is in writing and anyone who works on your tasks signs it.

5. Vetted people only

Anyone who touches your tasks has passed a scored trial on your rubric, signed the same NDA, and has their first batch double-reviewed. You know who is on the programme by name, and nobody is added without telling you.

6. One accountable owner

Every question above has the same answer to "who is responsible": me. Not a platform, not a trust-and-safety team, not a queue. If something goes wrong, you have a name and a same-day reply.

What I ask of you

Two things make this work from your side. Tell me the data-handling policy for the programme, even if it is a paragraph, so there is no guessing. And tell me who on your side can answer a question about it, so a policy question does not sit for a week while tasks wait.

Slide: NDA firstSlide: your platformSlide: never namedSlide: one owner
From the carousel version of this post.

Questions people ask

Do you sign an NDA before reviewing a task?
Yes, before any task, rubric, test suite or output is shared. A sanitised sample can be reviewed while legal review is in progress.
Will my programme be mentioned in your portfolio?
No. Clients and programmes are never named in public. Published numbers are aggregate and carry no client identification.
Do you use AI tools on client tasks?
Only with written approval under the client's stated policy. If no policy is stated, the default is none, and everyone working on the programme signs that policy.

If you're a founder

Your product idea gets the same treatment: nothing from a teardown or a sprint appears publicly without your written yes, and case studies are anonymised unless you want your name on them.

If you run an AI lab or data team

If your security questionnaire has questions this post does not answer, send it. I will answer in writing, then review one task free under NDA.

Sources

  1. Own policy, as applied on every programme
  2. 404 Media, contractors dismissed for using AI on training work, 22 Sep 2026

Free first step

Get a free written review of one task.
No call needed, no commitment.

Send one task, rubric or test suite. NDA on request, or a sanitised sample. Written review within 48 hours.

Send a task for review

Free · Written review in 48 hours · No commitment

Prefer email? prashantkumarsingh.work@gmail.com