Atlassian Can Use Jira And Confluence Data For AI After August 17. Can You Opt Out?

August 2, 2026

Starting August 17, Atlassian can use eligible Jira, Confluence, Rovo, and JSM data to improve AI across customers. Opt-out rights depend on plan and data type.

Yes. Starting August 17, 2026, Atlassian says it can use eligible Jira, Confluence, Jira Service Management, Rovo, and related platform data to improve apps and AI experiences for all customers. Every plan can turn off in-app content contribution, but only Enterprise organizations can turn off metadata contribution. The defaults also vary by plan.

Last updated: August 2, 2026

The Short Answer

Atlassian's change is real, but the phrase "Atlassian is training AI on all your Jira data" is too broad.

The official data-contribution FAQ separates two categories:

Data category Examples Free default Standard default Premium default Enterprise default Can an admin turn it off?
Metadata Content attributes plus common patterns from search, Rovo Chat, and configuration data On On On On Enterprise only
In-app data Confluence page titles and content; Jira titles, descriptions, and comments; custom workflows and statuses On On Off Off Yes, on every plan

Atlassian says it de-identifies and aggregates contributed data before using it across customers. It also says third-party hosted LLM providers such as OpenAI, Anthropic, and Google operate under zero-data-retention agreements and may not train their services on Atlassian customer data.

Atlassian itself may use contributed metadata to train its search model and fine-tune open-source models inside Atlassian infrastructure. That distinction matters: the policy is about Atlassian's cross-customer product and AI improvement, not permission for every outside model provider to keep Jira or Confluence content.

Who This Is For

This guide is for Atlassian organization admins, security teams, privacy teams, engineering leaders, and anyone whose work appears in Jira or Confluence.

The August 17 deadline deserves particular attention when an Atlassian site contains:

Individual users cannot make the organization-level choice on their own. Atlassian says the settings are controlled by organization admins in Atlassian Administration.

What Is Confirmed

1. The change takes effect on August 17, 2026

Atlassian's official data-practices page says the settings finished rolling out on May 19 and that the new use of eligible customer data begins August 17. Its updated AI terms, customer agreement, data-processing addendum, and privacy policy also take effect that day.

Today, Atlassian says it uses the relevant data to improve the experience inside a customer's own organization. After August 17, eligible contributed data can be used to improve apps and AI experiences for all customers.

The initial scope includes:

Atlassian says Loom, Trello, Bitbucket, Marketplace apps, and other apps without data-contribution settings are not initially included. The company says it will notify customers before adding settings for more apps.

2. "Metadata" includes more than ordinary technical telemetry

Atlassian defines metadata as content attributes plus common patterns.

Content attributes can include statistical or derived properties such as Jira story points, Confluence page complexity, task classifications, semantic-similarity scores, sprint dates, and service-level-agreement values.

Common patterns can be derived from:

The official data-type documentation says Atlassian omits low-frequency material that may be unique to one organization and uses recurring patterns seen across contributing customers.

This means "metadata" should not be interpreted as only timestamps, browser versions, or request counts. Some of it is derived from what users search, ask Rovo, and configure.

3. In-app data includes actual Jira and Confluence content

Atlassian's listed in-app data includes:

Free and Standard organizations have in-app contribution on by default. Premium and Enterprise organizations have it off by default. Organization admins on every plan can change the in-app setting.

The highest active plan in an Atlassian organization controls its defaults, including trials. If a company manages several Atlassian organizations, Atlassian says each organization must be reviewed separately.

4. Metadata and in-app content do not have the same opt-out rights

Every organization can turn off in-app data contribution. Metadata is different:

Atlassian lists exclusions for organizations using customer-managed keys or bring-your-own-key encryption, Government Cloud, Isolated Cloud, configured HIPAA compliance, or government-customer status. Educational institutions are not automatically excluded, even when government-run.

The practical lesson is simple: do not infer the setting from one product's plan or from a different Atlassian organization. Check the live organization-level page.

5. Atlassian distinguishes its own model improvement from third-party LLM training

Atlassian's FAQ says contributed customer data is not provided to third-party hosted LLM partners so those partners can train or improve their services. It names OpenAI, Anthropic, and Google and says those providers operate under strict zero-data-retention agreements.

Atlassian also says it may use contributed metadata to fine-tune open-source models that run inside Atlassian infrastructure. Its data-type documentation separately says metadata can train Atlassian's search model, while in-app data can improve recommendations, search relevance, acronym handling, and workflow next steps.

These are different claims:

  1. An outside hosted model provider does not get permission to train on Atlassian customer data.
  2. Atlassian may use eligible contributed data to improve Atlassian apps and AI across customers.
  3. Atlassian may train or fine-tune some models using contributed metadata under the data-contribution settings.

That separation is why a single yes-or-no question about "AI training" can produce a misleading answer.

6. Atlassian publishes removal and retraining commitments after opt-out

Atlassian says de-identified, customer-level aggregated data that is common across customers may be retained for up to seven years.

If an organization later opts out where the setting is available, or deletes an app or site, Atlassian says it will:

Those commitments are more specific than a generic promise to "respect your settings." They also show why changing the switch after August 17 is not the same as preventing contribution before it begins.

7. Turning off Atlassian AI is a separate control

Atlassian explicitly says data-contribution settings and AI activation settings are independent.

Turning off AI controls where AI experiences appear and whether users can invoke them. Data contribution controls whether eligible data can be used to improve Atlassian apps and AI experiences for all customers.

An admin should review both. Disabling one should not be assumed to configure the other.

What Is Still Unclear

The official documentation answers many operational questions, but several important details remain unclear to an outside reader.

How is de-identification tested?

Atlassian says it removes direct identifiers, aggregates data, omits rare material, and applies controls to prevent re-identification. The reviewed public pages do not publish the thresholds, attack tests, or residual-risk measurements used to decide that a phrase or pattern is common enough to retain.

De-identification reduces risk. It does not mean every derived topic, phrase, or workflow pattern is mathematically anonymous under every possible external dataset.

What happens to previously extracted common patterns after opt-out?

Atlassian says common, de-identified, customer-level aggregated data may remain for up to seven years. It also says it will stop identifying recurring patterns from an organization's prior contributions after opt-out.

The public wording does not clearly state whether every already-extracted common pattern is removed, whether it remains because it is no longer associated with one customer, or how the company proves that a retrained model no longer reflects a contribution that was blended with many others.

Which models are retrained, and on what schedule?

The promise to retrain models previously trained on opted-out data is unusually concrete. The public pages do not identify each affected model, publish a completion deadline beyond the dataset-removal windows, or describe how customers can verify completion.

Why does an older Atlassian support page still say customer data is never used to improve any model?

An Atlassian support article last updated November 6, 2025 still says Rovo and Atlassian Intelligence customer data is never used to train, fine-tune, or improve AI models or services. Atlassian's newer official data-contribution pages say the cross-customer change begins August 17, 2026 and describe model training and fine-tuning uses for contributed metadata.

The newer, date-specific documentation governs the upcoming change more directly. Atlassian should update or clearly time-bound the older support article so admins do not rely on a statement that becomes incomplete after August 17.

What This Change Does Not Prove

The policy change does not prove that:

Atlassian says it does not sell contributed metadata or in-app data, encrypts it in transit and at rest, respects configured data residency for contributed in-app data, and applies de-identification and aggregation before cross-customer use.

Those safeguards belong in the same explanation as the defaults, opt-out limits, and retention periods.

The MAP-IT Checklist Before August 17

Use this five-step review instead of treating the policy as a single switch.

M: Map every Atlassian organization

List each cloud organization, its Jira, Confluence, and Jira Service Management apps, its highest active plan, active trials, platform apps, and Teamwork Graph connectors. Settings do not automatically carry across separate organizations.

A: Audit the live defaults

Open Atlassian Administration > Security > Data contribution as an organization admin. Record the current metadata and in-app settings for each organization. Do not infer them from a purchase order or another site's plan.

P: Pause in-app contribution when the decision is not finished

Every plan can turn off in-app contribution. If security, privacy, legal, and product owners have not approved cross-customer use, turning it off before August 17 preserves the option to make a considered decision later.

I: Inspect the content and connector boundary

Sample what users actually store in Confluence pages, Jira comments, incident tickets, custom fields, Rovo Chat, searches, and connected sources. A low-risk project board and a security-incident workspace should not be evaluated as though they contain the same material.

T: Track the decision and later removal

Document who approved the setting, when it changed, which organizations it covered, and what evidence Atlassian provides if the organization opts out later. Track the 30-day in-app removal, 90-day content-attribute removal, and model-retraining commitments separately.

Where OpenVeil Fits

Atlassian is a workplace system of record and collaboration platform. OpenVeil is not a Jira or Confluence replacement, and it does not control what an organization stores in those services.

OpenVeil is a paid privacy-focused AI chat workspace with browser-local chat history and no normal server-side chat-history record for private chat sessions. That can be a useful narrower option for sensitive brainstorming or drafting when a user does not need a persistent project system, enterprise knowledge graph, or autonomous workplace agent.

The boundary still matters: active requests can be processed by OpenVeil and necessary AI, search, upload, hosting, security, billing, and infrastructure providers. OpenVeil is not fully offline, anonymous, zero-log, or a way to erase content already stored in Atlassian.

Before trusting any product's short privacy slogan, use the broader AI privacy-claim checklist. It helps separate training, retention, visible history, operational logs, provider processing, and deletion. You can also compare training opt-outs with chat retention and learn what browser-local AI chat history does and does not protect.

Frequently Asked Questions

Will Atlassian use Jira and Confluence data to train AI?

Atlassian says eligible contributed metadata and in-app data can be used to improve apps and AI experiences for all customers starting August 17, 2026. It specifically says metadata can train its search model and may fine-tune open-source models inside Atlassian infrastructure. The public pages describe in-app data as improving recommendations, search, and workflows, but do not say every contributed page or issue directly trains a generative foundation model.

Can Free or Standard customers opt out?

They can turn off in-app data contribution. They cannot turn off metadata contribution through this control. Free and Standard organizations have both categories on by default.

Can Premium customers opt out?

Premium organizations have in-app data off by default and can change that setting. Metadata is on and cannot be turned off through the data-contribution control.

Can Enterprise customers opt out?

Yes. Enterprise organization admins can turn off both in-app data and metadata contribution. Both are initially configured with metadata on and in-app data off, subject to organization-specific exclusions or settings.

Does turning off Rovo or Atlassian AI stop data contribution?

Do not assume it does. Atlassian says AI activation and data contribution are independent settings. Review both controls.

Does Atlassian let OpenAI, Anthropic, or Google train on this data?

Atlassian says no. It says third-party hosted LLM partners operate under zero-data-retention agreements and may not use Atlassian customer metadata or in-app data to train or improve their services.

Does opting out delete previously contributed data immediately?

No. Atlassian publishes separate windows: up to 30 days to remove corresponding in-app data from improvement datasets and up to 90 days for corresponding content attributes. It also says it will retrain models previously trained on the data and stop finding new recurring patterns from prior contributions.

Is the change a data breach?

No breach is established by the policy update. It is a planned change in permitted cross-customer data use, paired with organization-level settings and safeguards.

The Bottom Line

Atlassian's August 17 change is not a blanket handoff of every Jira and Confluence page to outside AI companies. It is also not merely a harmless telemetry update.

Eligible organizations can contribute actual in-app content plus derived metadata so Atlassian can improve apps and AI across customers. In-app content can be turned off on every plan. Metadata can be turned off only on Enterprise, although certain regulated, government, isolated-cloud, and customer-managed-key organizations are excluded.

Admins should check the live settings now, document the decision, and treat model training, cross-customer improvement, third-party processing, retention, and deletion as separate questions.

Sources

When privacy, account control, uploads, and search matter, OpenVeil gives you a private AI workspace designed for that job.