Atlassian auto-enrolled 300,000 customers into AI training. Starting August 17, 2026. Jira issue bodies, Confluence pages, comments, custom workflow names, sprint dates, story points, SLA numbers.
- Free: all of it by default.
- Standard: all of it by default.
- Premium: metadata only, no opt-out.
- Enterprise: off by default.
The pricing tier became the privacy tier.
- Retention: 7 years.
- Dataset removal after opt-out: 30 days.
- Model retraining: 90 days.
The reversal window is longer than most quarterly plans.
To their credit, Atlassian published 4 months in advance. Most vendors give 30 days. The 90-day retraining commitment is real work.
The problem isn't the disclosure. It's that three of four subscription tiers can't actually opt out of the metadata collection. That makes the disclosure a courtesy, not a choice.
The Jira I touch is all corporate work, and those clients are on Enterprise. So this policy doesn't hit me directly. What it does hit: every team I've seen pick Linear or Notion instead. Those tools don't have an Enterprise tier with a privacy moat (yet). The Atlassian tier gradient is the template. The next vendor copies it.
Every SaaS ToS is a data-residency decision you already signed. You didn't get opted in on August 17. You got opted in the day you signed up. Atlassian just told you about a consent they already had.
Your SaaS pricing tier is now your privacy tier. The data boundary is the check you write, not the checkbox you tick.
Link in comments.