Skip to main content

Effective date: 31 Mar 2026
Last updated: 30 Jul 2026

This policy explains how AOBRAIN SYSTEMS SL uses AI model infrastructure for EpicStory in Jira Cloud, how data is handled in AI workflows, and what controls apply in the current production release.

This policy is an operational disclosure. It is not legal advice.

1. Scope

This policy applies to EpicStory AI generation in Jira Cloud, including:

  • Epic initial generation;
  • Project initial generation;
  • full regeneration of an active Project draft;
  • Forge resolver, asynchronous queue, and Realtime delivery flows; and
  • operational signals related to AI calls.

Issue-level or selection-aware partial refinement and reusable Project sources are not available in version 7.

2. Current model provider posture

EpicStory uses Atlassian-hosted Forge LLM for production AI processing. The current model shown by EpicStory is Claude Haiku 4.5. Atlassian controls Forge LLM model availability, so the model may change when Atlassian changes the available model set.

The production path is:

  1. Jira provides permitted Epic or Project context to the Forge app.
  2. EpicStory’s Forge resolver and asynchronous consumer prepare generation stages.
  3. Forge LLM processes the request and returns generated text.
  4. EpicStory maps the result into an editable Jira draft.
  5. A user reviews, edits, selects, and deliberately publishes approved work.

Production has no external AI or observability egress configured by AOBRAIN.

Forge-hosted persistent storage supports Atlassian data residency controls. Transient Forge compute and model routing may occur elsewhere for platform reliability. EpicStory does not guarantee that inference occurs in a selected region or the same region as the Jira site.

3. Data sent for AI processing

Depending on the requested flow, AI input may include:

  • Jira Epic or Project text needed to generate a draft;
  • the user’s generation instruction;
  • selected Jira context permitted by the app scopes; and
  • an active Project draft when the user requests full Project regeneration.

EpicStory is designed to submit only the text needed for the requested generation. The current Forge LLM flow is text-only.

4. Storage and retention

Jira input and generated model output are transient generation data. They are passed through the active Forge generation flow and returned to the user as an editable draft; prompts and generated Jira content are not intentionally written to application logs.

Forge-hosted storage is used for operational jobs, identifiers, credit ledgers, usage records, and application state. These records are governed by bounded application lifecycle controls and Atlassian’s platform lifecycle. Detailed retention can vary by data class and operational need.

5. Training, moderation, and safeguards

AOBRAIN does not intentionally use customer Jira content to train public AI models.

Forge LLM is operated by Atlassian and is subject to Atlassian’s platform terms, safeguards, and moderation controls. Model output can still contain mistakes, omissions, unsupported assumptions, or unsuitable suggestions.

6. Human review and output limitations

All generated content is presented as an AI draft. Customers are responsible for reviewing and editing it before publishing or relying on it for delivery, security, legal, or compliance work.

EpicStory is not an autonomous decision-maker, security reviewer, legal adviser, or compliance reviewer. Publishing remains a deliberate user action.

7. Logging, metrics, and usage accounting

EpicStory uses Forge-native observability:

  • content-free Forge application logs;
  • built-in invocation and Jira API metrics;
  • sampled Forge custom counters; and
  • Atlassian Usage and charges reporting.

Application logs and custom metrics are designed to contain technical fields such as flow, stage, model, token counts, duration, finish reason, warning count, and safe error category. They are not intended to contain prompts, Jira descriptions, generated Jira content, account IDs, user IDs, or tenant identifiers.

Custom metrics may be sampled and are operational indicators, not billing records. Atlassian Usage and charges is the billing source of truth.

9. Contact

For AI data handling or privacy questions: