ROVO is using public or private LLM ? Is there any scenario where our data go outside Atlassian ?

[Miguel Velazquez]

Do Atlassian sign a contract for data security ? What happens in case of data breach in future ?

[Tomislav Tobijas]

Hey @Miguel Velazquez ,

Some of these things are answered here: Data contribution FAQs 
But generally, a lot of stuff can be found in Atlassian Trust Center 🛡️

If you still do have specific security questions, you can reach out Atlassian Support > select Security concern (e.g., vulnerability, suspicious activity, exposed data) and fill out the required info.

Cheers,
Tobi

[Sami Shaik]

Hello @Miguel Velazquez , welcome, and these are exactly the questions to ask before enabling Rovo, so here are the answers as Atlassian documents them today, with the one setting that changes the picture.

Which LLM: both, and it is not one model. Rovo routes across a mix: open-source models Atlassian hosts itself (Llama, Phi, Mixtral families) plus third-party hosted models (OpenAI GPT, Anthropic Claude, Google Gemini), selected automatically per task. You cannot pin a specific model today.

Does data leave Atlassian: by default, yes, in a narrow and contractually bounded way. For a Rovo request, the prompt, the response, and the relevant context (only content the requesting user is already allowed to see, plus any file they upload) may be sent to one of those third-party providers to generate the answer. Atlassian's stated terms with those providers are zero data retention: they do not store your inputs or outputs and do not use them to train or improve their models. Atlassian also states customer data is not used to train Atlassian's own models (it may use metadata to tune open-source models that run inside Atlassian's infrastructure). Usage telemetry stays with Atlassian and is not sent to the LLM providers.

If "outside Atlassian" is a hard no for you, there is a switch. Organization admins can enable Atlassian-hosted LLMs at the org level (Atlassian Administration). With it on, Rovo features use only models hosted inside the Atlassian Cloud boundary, and prompts and context are not sent to external providers at all. Two things to know before you flip it: it applies to the whole organization, not per site, and those hosted models currently run in a US data centre, which matters if you have residency requirements.

Contract and breach: the commitments above are not informal; they live in your Atlassian Cloud agreement, the Data Processing Addendum, and Atlassian's Trust Center documentation, and breach notification obligations are defined there. For a formal security review, ask your Atlassian account team for the current DPA and the Rovo-specific data usage documentation rather than relying on any of us here, and check the AI transparency pages for the model list, which does change.

Short version: default Rovo sends bounded, permission-scoped context to zero-retention third-party models; Atlassian-hosted LLMs keeps everything inside; both modes never train on your data.