Atlassian enables default data collection to train AI

Original Article ↗ Hacker News Discussion ↗

Atlassian is enabling default collection of “in-app data” from Jira, Confluence and related products to train AI models, with opt-out controls scheduled to appear only months before enforcement in August 2026. Commenters highlight serious concerns about sensitive corporate, customer and even regulated data being repurposed in this way, the vague distinction between “metadata” and content, and the inability to currently disable collection. Many see the move as part of a broader trend in SaaS toward opt-out-by-default AI training, prompting renewed interest in self-hosted or privacy-preserving alternatives and speculation about a link to rumored acquisition talks with AI firms.

Scope of Atlassian’s New Data Collection

  • By default, all free and paid customers are being opted in to contribute “in‑app data” for AI training.
  • “In‑app data” is described as user-generated content such as Confluence page titles/bodies, Jira issue titles/descriptions/comments, and even custom emoji, status, and workflow names.
  • Atlassian also defines “metadata” broadly as derived “content attributes” (e.g., page complexity, story points) and “common patterns” (frequent phrases, keywords, prompt topics), which many commenters argue is effectively still content.
  • Data residency (pinning data to a region) does not exempt customers from this data use.

Opt-Out Mechanism and Timeline

  • Many admins report that the documented “Data contribution” setting is currently missing from their instances.
  • Email communications say org-level opt-out settings will appear gradually and be available by May 19, 2026, with collection starting August 17, 2026.
  • Some interpret the delay and UI absence as intentional friction; others simply note it as a rollout issue.
  • A cited statement implies that if you terminate now, the new data-contribution controls don’t apply yet, which some see as preventing calm evaluation.

Security, Confidentiality, and Legal Concerns

  • Strong concern about highly sensitive content in Jira/Confluence (customer data, embargoed vulnerabilities, pharma investigations, health-related information) being used to train models and possibly leaking via AI outputs.
  • Questions raised about trade secrets, NDAs, and whether this undermines “reasonable efforts” to protect confidential information.
  • Government/HIPAA carve-outs are noted; some ask why trade secrets are not similarly carved out.
  • Whether Bitbucket repo content or Loom videos are included is unclear; policy wording is seen as vague.
  • Some expect little practical enforcement against violations.

Product Quality and Corporate Behavior

  • Numerous complaints about Jira/Bitbucket/Confluence reliability: broken or random search, desyncs, bugs in boards and navigation, poor input fields, AI features that don’t work, and difficult cancellation flows.
  • Explanations suggested: feature-chasing, technical debt, weak engineering, org churn, and cloud-only focus after dropping self-hosted editions; also broader “enshitification” and shareholder pressure.
  • A minority view calls this a rational business move that won’t change unless revenue is affected.

User Reactions and Alternatives

  • Some vow to leave Atlassian (“stop using this product” toggle) and migrate to GitLab, Linear, or self-hosted tools, citing existing export paths and migration scripts.
  • Others note that Atlassian is deeply embedded in workflows, or constrained by customer/regulatory rules, making migration hard.
  • Several argue that many SaaS vendors (e.g., developer tools, design tools) already default to training on customer data; the safest path is self-hosting.
  • There is interest in local-first, peer-to-peer replacements and open-source alternatives like self-hosted Confluence-like tools, but concern about operational burden and maintenance.

Rumored Acquisition and Motives

  • A rumor circulates that an AI company is in talks to buy Atlassian, presumably for its rich business-task dataset; some see the data policy as aligning with that.
  • Others dismiss this as unverified speculation or possible stock-pump chatter; no consensus is reached.