Starting August 17, 2026, Atlassian will use customer cloud product data from Jira, Confluence, and Jira Service Management to train its Rovo and Rovo Dev AI. If your team runs on the Free or Standard tier, you cannot fully opt out: metadata collection stays on, and only in-app data collection can be turned off. Enterprise customers get it off by default. Every other customer needs to open the admin panel and flip a switch before the cutover, or the default takes effect for them.
The change touches roughly 300,000 customers worldwide. Below is exactly what shifts, who controls it, and the one setting to check today. There is also a quieter risk that most coverage skips, and it is the one worth the most attention.
What changes on August 17, 2026
On that date, Atlassian begins feeding data from your cloud products into the training pipeline for Rovo and Rovo Dev, its AI assistants. This is not a new feature you switch on. It is a default policy change applied across the customer base, so the burden is on you to opt out rather than opt in.
Two categories of data are in play. The first is in-app data: the actual content inside your tickets, pages, and requests. The second is metadata: signals about how that content is used and structured. The distinction matters because your ability to say no depends on both the data type and your subscription tier. Most teams read a headline about AI training and assume there is a single opt-out button. There is not. There is a content lever and a metadata lever, and they behave differently.
The framing also matters. This is a default-on rollout, not a beta you can ignore until you are curious. When a policy ships as opt-out rather than opt-in, silence counts as consent. A team that reads no admin emails, changes nothing, and keeps working exactly as before will be contributing data on August 17 by simple inaction. The only way to end up on the other side is a deliberate change made before that date.
- Products affected: Jira, Confluence, and Jira Service Management on Atlassian Cloud.
- AI systems trained: Rovo and Rovo Dev.
- Effective date: August 17, 2026.
- Scale: about 300,000 customers worldwide.
The action item is narrow and the same for everyone who can act: open Atlassian Administration, go to Security, then Data contribution, and disable in-app data collection before August 17.
Who can opt out, and who cannot
Your tier decides how much control you have. The sharpest line is between Free/Standard and Enterprise. Free and Standard customers cannot opt out of metadata collection at all. Enterprise customers start with both metadata and in-app collection off by default, so they inherit the safest position without touching a setting.
In-app data collection is the one lever available to all customers. That is the setting in the Data contribution panel. Turning it off keeps the content of your tickets and pages out of the training set. Metadata is a different story, and for the lower tiers that switch simply is not offered.
It helps to be concrete about what metadata can carry. Even without the body text of a ticket, patterns around it describe a great deal: how many issues a project generates, how long they sit before resolution, which fields your team fills in, how pages link to each other, and the rhythm of when work happens. None of that is the secret prose inside a Confluence page, but in aggregate it still describes how your organization operates. For Free and Standard customers, that layer is not something you can switch off.
Enterprise sits at the other end. Both metadata and in-app collection are off by default, so an Enterprise organization that changes nothing still starts in the protected position. That inversion of the default is the single clearest reason the tier line matters here. The same inaction that opts a Free team in leaves an Enterprise team out.
The practical takeaway: if you are on Free or Standard, disabling in-app data collection is the most you can do. Metadata about how your team works will still be collected. Only Enterprise gives you a fully-off starting position.
Opt-out ability by tier
Here is the control you have on each tier, based on the confirmed facts. The states below match Atlassian's tier breakdown; still, confirm them in your own admin panel rather than assume a screenshot from a blog post matches your account.
| Control | Free | Standard | Premium | Enterprise |
|---|---|---|---|---|
| Opt out of metadata collection | No | No | No | Off by default |
| Disable in-app data collection | Yes | Yes | Yes | Off by default |
| Default state before you act | Metadata on; in-app on unless disabled | Metadata on; in-app on unless disabled | Metadata on; in-app off by default | Both off by default |
Premium sits between the extremes: metadata collection is on and cannot be turned off, the same as Free and Standard, while in-app data collection is off by default. Even so, confirm your own account's Data contribution settings directly rather than trusting a screenshot from a blog post.
How to turn it off, step by step
All customers can disable in-app data collection from one place. You need admin rights on the organization. Do this before August 17 so the change is live before the cutover, not after.
- Sign in to Atlassian Administration for your organization.
- Open the Security section.
- Select Data contribution.
- Disable in-app data collection.
- Save, then confirm the setting reads as off for each affected product.
One caution on scope. Disabling in-app data collection is the setting confirmed to be available to all customers, and it governs the content of your tickets and pages. It does not, on its own, remove metadata collection for tiers where metadata cannot be opted out. Read the labels in your own panel carefully rather than assuming one switch covers everything. If a control is not present for your tier, its absence is the answer, and the honest move is to plan around it.
If your organization has multiple Atlassian sites or products, check the Data contribution setting for each one. A single toggle in one workspace does not guarantee coverage across every site your company owns.
The risk most guides won't tell you about
Here is what most guides won't tell you: the exposure that matters is not the AI answers. It is that your team's private tickets, decisions, and design debates become training signal once they are in. Atlassian's own terms give a partial recall path: after you opt out or delete the apps, it removes the corresponding in-app data within 30 days and retrains any previously affected models within 90 days. But that is a cleanup window measured in months, not an instant undo, and it only helps if you opt out at all. Collected data can otherwise be retained for up to seven years.
Think about what actually lives in a mature Jira and Confluence instance. Postmortems naming what broke and who caught it. Roadmap arguments that never shipped. Security tickets describing exactly how a flaw was fixed. Salary-adjacent HR workflows in Jira Service Management. That corpus is the institutional memory of the company, and it was written with the assumption that it stays inside the tenant.
None of that content was authored with a training pipeline in mind. Engineers write blunt root-cause notes because they trust the audience. Product managers record why a feature was killed because the reasoning matters later. A support agent pastes reproduction steps into a request without pausing to consider a downstream use. The value of that writing comes precisely from its candor, and candor is exactly what you do not want reshaped into a general model outside your control.
The industry-wide pattern is that AI training increasingly runs on an opt-out default, which puts the timing burden on the customer. Miss the window, and the default decides for you. That is why the calendar date, not the toggle, is the real deadline.
There is a second-order effect that engineering leads underrate. Once a model has been trained on a corpus, the influence of that corpus is diffuse. It does not sit in a row you can delete. It is spread across weights in ways no admin panel exposes. So the mental model of privacy as a database you can purge does not map cleanly onto training data. The clean boundary you have today is before the cutover, while the choice is still yours to make.
This is also why the exposure scales with how well your team documents. The most disciplined teams, the ones with rich Confluence spaces and detailed Jira histories, have the most to contribute and therefore the most to weigh. Good documentation habits, ironically, raise the stakes of a default-on training policy. That is not a reason to document less. It is a reason to be deliberate about where the deepest context lives, and to review the Data contribution controls under Atlassian Administration before the August 17 cutover rather than after it.
What engineering teams and PMs should do this week
Treat this as a two-track task. Track one is the immediate toggle. Track two is a longer look at where your team's context actually lives and who can repurpose it. The first is done in an afternoon. The second is the conversation that keeps this from repeating the next time a vendor changes a default.
For smaller teams and SMBs on Free or Standard, the calculus is tighter. You have the least ability to opt out of metadata and often the least admin bandwidth to track policy emails. A simple internal note, one owner, one deadline, one screenshot of the confirmed setting, does more than a long policy discussion. Assign the toggle to a named person and mark August 17 as a hard date, not a someday.
- Confirm your tier. It determines how much you can actually opt out of.
- Disable in-app data collection in Data contribution before August 17, on every site.
- If you are on Free or Standard, document that metadata collection cannot be disabled and factor that into what your team writes in tickets.
- For Enterprise, verify the default-off state rather than assuming it; confirm no admin flipped it on.
- Review what sensitive context sits in Jira and Confluence that you would not want in a vendor's training set.
Keep your institutional memory portable and yours
This policy change surfaces a structural problem: the decisions and context your team depends on are locked inside a vendor that can change the terms and repurpose that data on a schedule you did not set. When the platform owns the memory, the platform sets the rules for it.
An external memory layer flips that arrangement. MemX keeps your team's decisions and working context in a store you control, portable across the AI tools you use rather than trapped in one vendor's training pipeline. It is private by architecture, with per-user isolation, encryption at rest, and on-device options, so the context you rely on stays context you own. That does not replace Jira or Confluence. It means the institutional knowledge you care most about is not only wherever a vendor's default policy happens to point next.
01When does Atlassian start training AI on my data?
August 17, 2026. On that date Atlassian begins using cloud product data from Jira, Confluence, and Jira Service Management to train its Rovo and Rovo Dev AI systems, applied as a default across the customer base.
02Can I opt out of Atlassian AI training on the free plan?
Partly. Free and Standard customers can disable in-app data collection but cannot opt out of metadata collection. Only Enterprise starts with both metadata and in-app collection off by default.
03How do I turn off Atlassian data contribution?
In Atlassian Administration, open Security, then Data contribution, and disable in-app data collection. Do this before August 17, 2026, and check each site if your organization runs more than one.
04How many customers does this affect?
About 300,000 customers worldwide use the affected Atlassian cloud products, so this default policy change reaches a large share of teams running Jira, Confluence, and Jira Service Management.
05Does opting out remove data already used for training?
Partly. Atlassian says that after you opt out or delete the apps, it removes the in-app data within 30 days and retrains affected models within 90 days. That is a months-long cleanup, not an instant undo, so acting before August 17 still matters most.
