Artificial intelligence is helping software engineers code faster. While it’s a boon for many, hackers are also using it for malicious purposes. According to a new report, the hackers have used SpaceX’s AI coding assistant, Cursor, to help break into a Belgian chemical company and at least six other businesses. It was recently uncovered after cybersecurity startup Gambit Security found an internet-exposed server linked to a new ransomware group called Aur0ra. The investigators stated that they reviewed 28 chat sessions between the hackers and a Cursor AI agent. The conversations showed the attackers using false claims about running a security test to persuade the AI to provide help with activities such as stealing login details and taking over important accounts.
Reuters, citing Gambit Security, recently reported that the hackers used Cursor’s AI agent to carry out hundreds of harmful tasks. According to the report, the attackers asked the AI to find administrator accounts and working passwords. The shocking part was that in one case the agent helped after the hackers gained access to a company’s VPN. If that was not already enough, the tool also provided suggestions for breaking password hashes.
Also read: Samsung Galaxy S26 FE vs Pixel 10a vs iPhone 17e: Camera, battery, specs, price and more compared
The chat logs reviewed by Reuters covered activity between April 8 and May 21. The media outlet claimed that the targeted organisations included Belgian hygiene and cleaning products maker Christeyns, German garage door manufacturer Teckentrup and the Scotland-based Helideck Certification Agency. Other than that, an Argentine pharmaceutical distributor, an Italian manufacturer and Bayou Title, a Louisiana-based title insurance company, were also part of the alleged hack.
The most shocking part was how the attackers managed to get around the AI’s safety measures. Gambit said Cursor’s agent refused some harmful requests, but the hackers repeatedly restarted conversations and claimed that their activities were part of a legal simulation or security test.
Moreover, the logs also showed AI accepting these explanations and then providing technical assistance. At one point, it reportedly described a suggested attack as having a ‘VERY HIGH’ chance of success.
Also read: Samsung Galaxy S26 FE launches with One UI 9 and 4900mAh battery: Check price, specs and more
Gambit also added that the AI agent was powered by Anthropic’s Claude Sonnet 4.5. The cybersecurity company believes the AI gave the attackers a significant speed advantage by reducing the amount of work they had to perform manually.
However, Reuters added that they were unable to independently confirm how much AI directly contributed to each intrusion or whether every attack resulted in stolen data or ransom demands.