Hidden PDF text hijacks Atlassian AI to leak your Jira and Confluence data

A security firm has found that Atlassian's AI assistant is vulnerable to a prompt-injection attack: instructions hidden in what appears to be a blank PDF can silently instruct it to forward Jira tickets and Confluence documents to an attacker. The exploit operates without the user's knowledge, as the malicious commands are invisible to the naked eye. Atlassian has not publicly commented on the finding.

Similar stories

💻 Technology 'Context bombing' tricks malicious AI hacking agents into shutting down Wired · 23d ago 💻 Technology House Democrats demand tech CEOs testify under oath over AI-driven cybersecurity breaches Gizmodo · 4h ago 💻 Technology CrowdStrike and Palo Alto hit record highs after Black Hat AI threat warnings CNBC · 5h ago