Jira Tickets: New Vulnerabilities in ‘Living Off AI’ Attack Proof of Concept

Cato Networks has raised concerns about potential security vulnerabilities in Atlassian’s Jira Service Management (JSM) due to AI prompt injection risks. In a recent blog post, Cato showcased a proof-of-concept (PoC) demonstrating how support tickets could be exploited to manipulate AI systems, specifically those integrated with Atlassian’s Model Context Protocol (MCP). The MCP allows AI models like Anthropic’s Claude to assist in managing Jira and Confluence workflows. Cato’s demonstration illustrated that an attacker could craft a support ticket that prompts Claude to leak internal data, posing significant risks when the AI handles unverified external inputs. The threat is particularly pronounced as some JSM portals allow ticket submissions without authentication. Atlassian has stated that customer data security is a priority and is exploring enhanced controls to mitigate these risks, urging users to carefully evaluate their MCP implementations. Similar AI-related vulnerabilities have emerged in other platforms, indicating a broader pattern of concern.

Source link