Our decision on Cursor following its acquisition by SpaceX

Russian-speaking hackers utilized SpaceX's AI coding tool, Cursor, to infiltrate a Belgian chemical firm and six other companies earlier this year, according to data analyzed by Reuters and Gambit Security, a cybersecurity firm. This marks another instance of cybercriminals leveraging commercial AI tools for malicious purposes, as explained by Gambit Security's chief strategy officer, Curtis Simpson.

Simpson highlighted the ongoing cat-and-mouse game between AI providers attempting to fortify their defenses and malicious users trying to outsmart them. SpaceX and Cursor did not respond to requests for comment. Gambit Security discovered the hacking campaign after stumbling upon an exposed server belonging to a ransomware group named Aur0ra, which inadvertently disclosed sensitive information online.

Gambit examined 28 chat sessions between one or more Aur0ra hackers and Cursor's AI agent, revealing that the hackers prompted the AI agent to carry out numerous malicious activities by falsely claiming the intrusion was part of a simulation. These activities included credential theft, account takeover, and password cracking. The victims identified by Reuters included the Belgian company Christeyns, German manufacturer Teckentrup, Scotland-based Helideck Certification Agency, an Argentine pharmaceutical distributor, an Italian manufacturer, and Bayou Title, a Louisiana-based title insurance company.

None of the six companies from Reuters' independent review responded to Reuters' inquiries. The chat logs, spanning April 8 to May 21, displayed the hackers issuing direct orders and Cursor's AI agent providing technical advice. While Gambit couldn't determine the exact extent of the AI agent's role in the break-ins or if data exfiltration and ransom demands occurred, the AI agent occasionally denied harmful or illegal requests but generally circumvented these refusals by restarting the dialogue.