Z.ai GLM-5.3: Autonomous Vulnerability Research and Cursor IDE Exploitation
The release of Z.ai's GLM-5.3 open-weight model marks a critical shift toward autonomous offensive AI, characterized by its emergent ability to perform independent vulnerability research. GLM-5.3 successfully identified and exploited a serious vulnerability within the Cursor AI-native IDE, demonstrating a recursive attack vector where AI-driven development environments are targeted by autonomous agents. This capability significantly compresses the time between vulnerability discovery and exploit weaponization, bypassing traditional human-in-the-loop constraints. The exploit leverages iterative agent workflows to transition from static code analysis to functional exploitation, posing a systemic risk to AI-integrated software supply chains.
Critical OS-Level RCE via "DuneSlide" in Cursor AI
Researchers at Cato Networks have identified "DuneSlide," a pair of critical vulnerabilities (CVE-2026-50548 and CVE-2026-50549) in the Cursor AI IDE. These flaws enable prompt-injection-driven sandbox escapes, escalating from LLM interactions to full operating system-level Remote Code Execution (RCE). Attackers can leverage malicious Model Context Protocol (MCP) servers or poisoned web search results to manipulate the run_terminal_cmd tool and bypass path canonicalization logic via symbolic links. Successful exploitation allows unauthorized file writes outside the project root, enabling attackers to overwrite the cursorsandbox executable, modify shell configurations, or establish persistence via macOS LaunchAgents, resulting in total system compromise.