A vulnerability called CurXecute in the AI-powered code editor Cursor allows attackers to execute remote code with developer privileges. This vulnerability, identified as CVE-2025-54135, can be exploited by feeding the AI agent a malicious prompt, leading to ransomware and data theft incidents. The Cursor IDE relies on AI agents to help developers code faster, but this vulnerability exposes it to external, untrusted data that can compromise the agent's control flow. A hacker could hijack the agent's session and privileges to act on behalf of the user, potentially leading to significant data leaks and other malicious activities. The vulnerability was reported and patched, with the latest version of Cursor addressing the issue.

A critical security vulnerability in Cursor, an AI-powered code editor for macOS, allows malicious software to bypass Apple’s privacy protections and access sensitive user data without authorization. The flaw, due to a misconfiguration in the Electron framework, enables attackers to execute arbitrary code with the app’s privacy permissions. This compromises the Transparency, Consent, and Control (TCC) framework, which protects user privacy on macOS systems. The vulnerability remains unpatched, posing a significant threat to users' sensitive information and potentially leading to data leaks and unauthorized access to hardware components.