Your team just connected Claude to Jira. An employee gives it a simple prompt: "summarize my week."
The recap looks great. It includes their sprint tickets and the bug due on Friday. But it also pulls in the performance review, a payroll approval, and notes from last week's leadership meeting in Confluence.
Nothing is broken. The employee has permission to access all of it. Claude simply retrieved everything those permissions allowed.
The problem is where that data ends up. It now sits in a chat history, outside your permission schemes and space restrictions. If someone shares that chat, or the AI account gets compromised, the data goes with it. With employee and payroll records, that's a privacy and compliance issue.
Your AI gets the same keys as your users
You've probably spent a lot of time getting permission schemes, project roles, and space restrictions right. Claude and ChatGPT inherit all of it.
When you connect them to Jira and Confluence, they work with the user's permissions. If a person can open a project or space, so can their AI assistant. For a lot of users, that includes HR, finance, legal, or security work.
Atlassian does give you some controls. You can allow or block Read, Write, and Search for each app. But those settings apply to the whole org. You can't let AI read the Engineering project and keep it out of HR.
So you're left with three options, and none of them are great:
- Limit AI for everyone.
- Remove people's access to sensitive projects.
- Let AI access everything the user can.
This isn't a theoretical risk. IBM's 2026 Cost of a Data Breach Report found that 92% of organizations hit by an AI-related breach lacked proper access controls for their AI systems.
How are you handling this in your organization?
Do you rely on existing Jira permissions, restrict AI access altogether, or use additional controls to protect sensitive projects?
