A GitHub user went looking for the model training opt-out setting, followed a helpful pointer from another community member, and came up empty. Their Privacy section only showed “Suggestions matching public code.” No toggle for training. Just a screenshot, a shrug, and the quiet implication that the setting either lives somewhere else or does not exist where the documentation says it should.
That exchange says more about the current state of AI data governance than any vendor blog post. The opt-out exists in principle. Whether you can actually find and click it is a separate question.
Why an SEO person cares about this at all
I run keyword research, competitive teardowns, content briefs, and client positioning docs through AI tools all day. Those prompts are not generic. They contain client names, unlaunched product lines, internal traffic numbers, and the specific reasoning behind why we are going after one query cluster instead of another. That reasoning is the actual work product. The deliverable is downstream.
So when a platform says it may use “interaction data — specifically inputs, outputs, code snippets, and associated context” to train and improve its models, I read that as my strategy notes becoming somebody’s training gradient. A Hacker News commenter called the distinction between “we read your private repo” and “we train on your Copilot interactions inside your private repo” unnecessary hair-splitting. I am inclined to agree. If the context window ate it, the context window ate it.
The current scorecard
Here is where things stand across the tools a lot of us have in our stack:
- Mistral — offers a real opt-out. Once confirmed, Mistral no longer uses your input or output data to train its models. Clean, explicit, done.
- GitLab — also offers an opt-out for training on your data.
- GitHub Copilot — may use private repo interaction data without a prior opt-out, and community reports suggest the setting is not always where users expect to find it. The Apr 24 deadline discussion on Hacker News tells you how much lead time people felt they had.
- Atlassian — starting August 17, 2026, will begin collecting customer metadata and in-app content from Jira, Confluence, and other cloud products to train its AI offerings, including Rovo and Rovo Dev.
- Vibe — has a documented opt-out path on the iOS and Android mobile apps.
Look at the Atlassian entry again, because that one hits agencies hardest. Confluence is where a lot of SEO teams keep content calendars, audit findings, migration plans, and client-facing strategy pages. Jira is where the tickets describing exactly what you are shipping and when live. “In-app content” is a broad phrase, and content is the whole job for us.
The legal side is not settled
Europe is circling the same problem from the rights-holder direction. Under the EU AI Act’s opt-out trend, it is not yet clear how a legally valid opt-out request would even be made. Organizations like the Dutch collecting society Pictoright, which handles pictures, and the French society Sacem are pushing on it. That ambiguity matters to anyone publishing content for a living. A reservation of rights that nobody can define procedurally is not much of a reservation.
Which means the practical answer, for now, sits in platform settings rather than in statute.
What I actually do about it
Nothing here is exotic. It is mostly hygiene.
- Audit every AI tool in the stack and write down, per tool, whether an opt-out exists and where it lives. Not “I think it does.” A screenshot or a settings path.
- Treat calendar dates as action items. August 17, 2026 for Atlassian is a ticket, not a fun fact.
- Check enterprise and team tiers separately from personal accounts. The controls often differ, and the admin who owns them may not be you.
- Assume the default is on until you confirm otherwise. GitHub’s situation is the argument for that assumption.
- Sanitize prompts for genuinely sensitive client work. Placeholder client names cost you nothing and remove the question entirely.
- Re-check after every privacy policy update. Settings get moved, renamed, and reset.
Where this leaves us
The opt-out question has a frustratingly accurate answer: sometimes, depending on the vendor, if you can locate the toggle. Mistral and GitLab show it can be a normal product feature. GitHub shows what happens when it is buried or absent. Atlassian has given the industry a date to prepare for, which is more courtesy than some.
For those of us whose competitive edge lives in documents and prompts, the useful posture is boring diligence. Read the notices, find the settings, keep a list. The tools are excellent. The defaults are the part you have to manage yourself.
🕒 Published: