Atlassian Is About to Train AI on Your Jira and Confluence Data
Starting August 17, 2026, Atlassian begins using customer metadata and in-app content to train AI features — on by default for most plans. Here's exactly where the opt-out switches live, and what they do and don't actually turn off.
If you're reading this before August 17, 2026, change your settings now. There's no grace period for waiting: Atlassian begins collecting data according to whatever your settings are on that date, and going forward from whenever you actually make the change. Read below for what changing the toggle does and doesn't undo.
Treat this as what it actually is: a change to a default setting that governs what leaves your tenancy is a governance decision, not a passive product update — and governance decisions have a named approver and a date, the same as any other change to what data leaves your organization. The question worth asking internally before you touch a toggle isn't just “how do I turn this off” — it's who in your organization reviewed and approved your current data contribution setting, and when? For most teams, as of this writing, the honest answer is nobody has, which is itself the finding.
Two Different Things Are Being Collected
Atlassian's policy — officially called data contribution — splits into two categories with different rules. Conflating them is exactly how you end up thinking you've opted out when you've only turned off half of it.
Metadata
De-identified, aggregated statistics derived from your content — not the content itself.
- •Readability and content-complexity scores
- •Task classifications (e.g. “sales work item”)
- •Semantic similarity scores between pages or issues
- •Story points and sprint end dates (Jira)
- •SLA metrics (Jira Service Management)
- •Search queries and results across customers
- •Rovo Chat prompts and responses
- •Custom field configurations
In-app data
The actual content your team wrote, before de-identification.
- •Confluence page titles and body content
- •Jira and JSM work item titles, descriptions, and comments
- •Custom emoji names
- •Custom status names
- •Custom workflow names
Both categories are de-identified (names and emails stripped) and aggregated before use, per Atlassian. Coverage today is limited to Jira, Confluence, Jira Service Management, and Atlassian's core platform apps (Rovo, Home, Teams, Projects, Assets, Goals, Analytics, Administration) — Marketplace apps, Bitbucket, and Loom don't have data contribution settings yet, which means there's currently no toggle for them at all, in either direction.
What Your Plan Actually Lets You Turn Off
The defaults — and whether you can change them at all — are set by the highest active plan in your Atlassian organization, not per-product.
| Plan | Metadata default | Metadata opt-out? | In-app data default | In-app opt-out? |
|---|---|---|---|---|
| Free | Always contributed | No | On by default | Yes |
| Standard | Always contributed | No | On by default | Yes |
| Premium | Always contributed | No | Off by default | Yes |
| Enterprise | On by default | Yes | Off by default | Yes |
In plain terms: on Free, Standard, and Premium, metadata is contributed no matter what you do — the only lever you get is in-app data. Only Enterprise gets a real all-off switch. Premium customers wanting to opt out of metadata specifically have asked Atlassian about this directly and been told it isn't available below Enterprise.
Atlassian confirmed this directly on its own community forum in July 2026. Asked whether Premium customers could opt out of metadata contribution without upgrading to Enterprise, an Atlassian Team member replied that, even with in-app data contribution switched off, Premium orgs still contribute “search queries and results, prompts and responses in Rovo Chat, configuration items you define (for example, names of custom fields in Jira)” — and that full exclusion is limited to “customers with at least one Enterprise plan or trial” plus the narrow compliance carve-outs (CMK/BYOK, Government Cloud, Isolated Cloud, HIPAA) covered above.
How to Change Your Settings
This is an organization-level setting, not a per-site or per-project one — you need Atlassian organization admin access, not just Jira or Confluence site-admin access.
- 1
Go to admin.atlassian.com and select your organization
If you manage more than one Atlassian organization, repeat every step below for each one — settings don't carry across organizations.
- 2
Open Security → Data contribution
This is the single screen that controls both metadata and in-app data contribution for the whole organization.
- 3
Turn off “In-app data contribution”
Available on every plan, including Free. This is the one lever every customer has.
- 4
On Enterprise, also turn off “Metadata contribution”
This toggle simply doesn't render on Free, Standard, or Premium — it isn't hidden or permission-gated, it's not offered at those tiers.
- 5
Optional: scope it instead of using the all-or-nothing switch
The same screen has “Include in-app data” and “Exclude in-app data” tables with tabs for Spaces, Atlassian apps, and Teamwork Graph connectors — useful if you want most spaces contributing but need to carve out a few confidential ones (or the reverse).
- 6
Expect a lag, not an instant wipe
Per Atlassian's own documentation, opting out removes in-app data from the datasets Atlassian uses to train its models within 30 days, removes content attributes within 90 days, and Atlassian states it will “retrain any models previously trained on that data.” What Atlassian doesn't publish is a verification method or a timeline for that retraining — so if your auditor wants evidence the removal actually happened, what you currently have is a vendor commitment, not an artifact you can hand them.
Per-Plan Quick Reference: Where the Switches Actually Live
Same screen for everyone — admin.atlassian.com → your org → Security → Data contribution — but what renders on it depends on your plan. This is what you'll actually see when you get there, by tier.
Free / Standard
What you'll see: One toggle: “In-app data contribution.” No “Metadata contribution” toggle renders at all — it isn’t hidden behind a lock icon, it simply isn’t part of the page at this tier.
What to do: Turn off in-app data. That’s the whole lever you have.
Premium
What you'll see: Same single “In-app data contribution” toggle, defaulted off already. Still no “Metadata contribution” toggle — the July 2026 Atlassian Community reply above confirms this is by design, not a bug.
What to do: Confirm in-app data contribution is off. Metadata (search queries, Rovo prompts, custom field names) keeps flowing regardless.
Enterprise
What you'll see: Both toggles render: “In-app data contribution” and “Metadata contribution,” plus the “Include/Exclude in-app data” tables (Spaces, Atlassian apps, Teamwork Graph connectors) for scoping instead of an all-or-nothing switch.
What to do: Turn off both toggles for a full opt-out, or use the include/exclude tables to carve out specific spaces or connectors.
CMK / Gov Cloud / Isolated Cloud
What you'll see: The Data contribution screen still exists, but these orgs are excluded automatically at any tier — the toggles are moot, not something you need to set.
What to do: Nothing to do. Verify the exclusion in your org's compliance/trust documentation if an auditor asks.
Atlassian's admin console changes layout without notice — if what you see doesn't match this, trust the screen in front of you and treat this table as a guide to what to look for, not a pixel-exact walkthrough.
For Compliance: What to Put in Your Vendor Register
If your organization maintains an AI-vendor inventory under ISO/IEC 42001 (the AI management system standard), or an ICT third-party register under DORA Article 28 (the EU Digital Operational Resilience Act's requirement to record contractual arrangements with ICT third-party providers), Atlassian's data contribution setting belongs in it as a line item, not a footnote. The fields worth capturing, based on what Atlassian does and doesn't publish above:
| Field | What to record |
|---|---|
| Vendor / service | Atlassian (Jira, Confluence, JSM, Rovo, Home, Teams, Projects, Assets, Goals, Analytics, Administration) |
| Plan tier at time of review | Free / Standard / Premium / Enterprise — determines which toggles are even available (see table above) |
| Data categories contributed | Metadata (always, below Enterprise) vs. in-app data (toggleable on every plan) |
| Opt-out exercised | Date the org admin changed the setting, and by whom |
| Stated removal timeline | 30 days (in-app data from training sets), 90 days (content attributes) — per Atlassian's own documentation, not independently verified |
| Retention of de-identified data | Up to 7 years even after opt-out, per Atlassian's published terms |
| Evidence of retraining | Atlassian states it will retrain affected models but publishes no verification method or completion date — record this as an open item, not a completed control |
| Next review date | Atlassian's AI and data-contribution policies have changed multiple times in 2026 — set a recurring review, not a one-time entry |
This is a starting template, not legal or compliance advice — confirm the specific fields your ISO/IEC 42001 or DORA Article 28 documentation requires with whoever owns that program at your organization.
What Opting Out Doesn't Do
Be precise about what the toggle actually covers before you tell your org this is handled.
Below Enterprise, metadata is contributed no matter what — story points, SLA values, search queries, and Rovo Chat prompts/responses are collected even with in-app data fully off.
De-identified, aggregated data can still be retained for up to seven years — in Atlassian's own words, data “de-identified and aggregated at a customer level such that it is no longer associated with individual customers… may be retained for up to seven years,” separate from the 30/90-day removal timeline for opted-out data.
Atlassian commits to retraining models already trained on your data once you opt out, but its published documentation doesn't say how fast, or how a customer would confirm it happened. If your compliance process needs evidence of removal, not just a stated commitment, that gap is currently yours to manage.
This setting doesn't reach Marketplace apps, Bitbucket, or Loom — those products don't have data contribution controls yet, so there's nothing to toggle either way.
“Data contribution” (training Atlassian's own in-house models) is a different promise from third-party AI training: Atlassian states customer data is never shared with outside AI vendors to train their models, but metadata can still be used to fine-tune Atlassian's own models inside its infrastructure, subject to this same setting.
If your organization runs multiple Atlassian organizations (common after M&A, or separate dev/prod orgs), each one has its own setting — there's no global override.
If You're Regulated: HIPAA, Government Cloud, or CMK
Organizations on customer-managed encryption keys (CMK/BYOK), Atlassian Government Cloud, or Atlassian Isolated Cloud are excluded from data contribution automatically, at any plan tier — nothing to configure.
HIPAA is the one that trips people up: the exclusion isn't automatic just because you're a healthcare organization on the free tier. You have to formally notify Atlassian and configure your organization for HIPAA compliance (a signed Business Associate Agreement, following Atlassian's HIPAA implementation guide) before the exclusion applies. One free-tier dental practice on the Atlassian Community forums reported the exclusion arriving by email only after completing that process — months after they first asked.
Related Reading
- What Atlassian's own data residency page doesn't cover — “AI data” can't be pinned to a region →
- Every Atlassian Cloud and Data Center price increase, 2023–2026, dated and sourced →
- The Data Center end-of-life countdown, and what it means depending on your status →
- What Jira, Confluence, and Data Center actually cost a 100-person team →
- The migration gotchas that lock teams into Jira — and a 2-minute lock-in audit →
- Atlassian kept one product self-managed. It's not Jira or Confluence →
- Your code can stay on-prem. Your build can't — the Bitbucket/Bamboo CI gap →
Settings Changed. Still Weighing Your Options?
Data control is one more variable in the platform decision, not the whole decision. If you're already sizing your Atlassian footprint, get a side-by-side cost comparison built from your actual numbers.
Size My InstanceSources
- Atlassian will collect Jira and Confluence data by default to train AI models — ghacks.net
- Understand data contribution — Atlassian Support
- Data contribution settings — Atlassian Support
- Data contribution settings availability — Atlassian Support
- What types of data does my organization contribute? — Atlassian Support
- Include or exclude in-app data — Atlassian Support
- Manage data contribution for your Atlassian organization — Atlassian Support
- Data practices built for responsible AI — Atlassian Trust Center
- Data Contributions Frequently Asked Questions — Atlassian Trust Center
- Rovo and Atlassian Intelligence: customer data is not used for third-party AI model training — Atlassian Support
- “Premium customers: can we opt out of metadata contribution without Enterprise?” — Atlassian Community
- “Premium customers: can we opt out of metadata contribution?” (with Atlassian Team reply, July 2026) — Atlassian Community
- “How can a free-tier healthcare organization opt out of Data Contribution / AI training?” — Atlassian Community
Compiled August 2026 from Atlassian's own support and trust center documentation, cross-checked against Atlassian Community threads for real-world admin experience. Atlassian's AI and data policies change; verify current settings and defaults at support.atlassian.com before relying on anything above.