Atlassian Now Trains Its AI on Your Work by Default — and Full Opt-Out Is an Enterprise Feature

If you run a team on Jira or Confluence, the deal changed on 17 August and the change was opt-out. From that date, by Atlassian’s own account, the content your team writes into its Cloud products — Confluence pages, Jira tickets, the descriptions and comments where the actual work lives — is used by default to train Rovo, Atlassian’s AI assistant. You were not asked to opt in. You were, at best, given a switch and left to find it.

Answer first, because the detail matters more than the outrage: there are two settings, and they are not equal. One governs your in-app data — the text itself. The other governs metadata — the derived signals about that text. On the Free, Standard and Premium plans you can turn off the content, but the metadata switch is greyed out; Atlassian’s support page reads, flatly, “You can’t change this setting.” The full off switch, the one that also stops metadata contribution, is available only on Enterprise. Privacy, in other words, is now a plan tier.

What actually changed, with the switches named

Atlassian’s data-contribution documentation lays out a matrix that is worth reading slowly, because the defaults are doing the heavy lifting. In-app data contribution defaults to on for Free and Standard customers and off for Premium and Enterprise. Every tier can toggle that one. Metadata contribution is a different story: it is on across the board and can only be switched off by Enterprise. So the customer contributing the most by default — content and metadata, both on, no ability to fully stop it — is the one on the cheapest plan who never opened the settings page.

The categories are broad. In-app data, per Atlassian’s materials, covers Confluence page titles and body text, Jira work-item titles, descriptions and comments, and custom status and workflow names. Metadata covers the derived layer: readability scores, task classifications (that a ticket is “sales work,” say), story points, sprint end dates, SLA values, and semantic-similarity measures drawn from the Teamwork Graph. None of that is your raw prose, but a great deal can be inferred from it — how your team works, what it’s working on, how fast, and how it’s all connected.

Opt-out is a real setting, and also a burden shifted onto you

To be fair, and we will be: Atlassian did ship an opt-out, it documented it, and the path is not buried in the way these things sometimes are. An org admin goes to Atlassian Administration, then Security, then Data contribution, and flips in-app data contribution off. That is more than some companies bother with, and it deserves acknowledging rather than skipping past.

But opt-out at the org level has a structural feature worth naming: it moves the work, and the risk of inaction, onto the customer. The admin has to know the change happened, know the setting exists, and act before the default does the contributing for them. Everyone in the workspace who is not the admin — the writer filing tickets, the engineer documenting a system — gets no switch at all; their data’s fate is decided a level up. And on the plans where metadata can’t be turned off, even a diligent admin who flips every switch they’re given is still contributing. The opt-out is genuine. It is also incomplete by tier, and that is the part that reads as anti-consumer.

The steel-man: why a company would do this

The good-faith case is not hard to state. Rovo competes in a market where the assistant that understands your workspace is more useful than a generic one, and the raw material for that understanding is your data. Atlassian has said the collection feeds product improvement, applied a distinction between derived metadata and actual content, and gave the more privacy-sensitive plans a gentler default — in-app content off for Premium and Enterprise. Retention is bounded on paper: opt out and in-app data leaves the improvement datasets within about 30 days, content attributes within about 90, with retraining to follow; aggregated, de-identified data may persist up to seven years. This is a more considered posture than a blanket “we train on everything, forever.”

Concede all of that, and the objection narrows to something specific and fair. It is not “a software company built an AI feature,” which is barely news. It is that a change with real privacy consequences was made the default, that the completeness of your control over it is metered by price, and that the people with the least protection are the ones paying the least. That is a choice about who bears the downside, and right now it points downward.

Where this sits in the pattern

We have watched this shape before from other directions. When a platform decides your content is training data by default, the specifics differ but the move rhymes: it happened when Twitch opted every streamer into training Amazon’s AI, and it is the quiet logic behind why nearly every AI product is so keen on your data in the first place. The workplace-tools version is arguably stickier, because you cannot casually migrate a decade of Jira history the way you can switch chatbots.

There is also a harder question lurking under Atlassian’s tidy retention timeline. Removing a document from a training set is not the same as removing its influence from a model that has already learned from it; that is the whole difficulty of machine unlearning — the subject of our companion explainer — and it is why “we’ll retrain” is a promise worth reading closely rather than taking on trust. Related, if the model memorises more than intended, is the ongoing risk that models can leak the data they were trained on.

What to actually do

The pro-consumer advice is boring and effective:

• Check the setting today. Atlassian Administration → Security → Data contribution. Turn off in-app data contribution if you don’t want your content used; on Enterprise, turn off metadata contribution too.
• Know your tier’s ceiling. On Free, Standard and Premium you cannot stop metadata contribution — factor that into what sensitive material you keep in a Cloud workspace at all.
• Tell the humans, not just the admin. The people writing the tickets and pages have no switch; the org that decides for them should at least decide out loud.
• Read the retention promise as a promise. “Removed within 30 days” and “retrained” are commitments to hold Atlassian to, and the seven-year aggregated-data window is the part that outlives your opt-out.

Atlassian is not uniquely villainous here; it is doing the thing the whole industry is drifting towards, and doing it with more documentation than most. The specific, fixable fault is the tiering. An opt-out that only becomes complete when you upgrade to Enterprise is an opt-out that says the quiet part plainly: your privacy is worth exactly as much as your contract. The fix is not complicated. Make the metadata switch available on every plan, and let the default be off. Until then, the setting is there — go and find it before it finds you.