Reuters: Russian-speaking hackers breached seven companies by tricking the AI agent in Cursor, the coding tool now owned by Elon Musk’s SpaceX, into thinking the attacks were a test

Russian-speaking hackers from Aur0ra, a new ransomware group, used the AI coding assistant built into Cursor — a code editor that Elon Musk’s SpaceX recently bought — to break into seven companies around the world, according to Reuters, citing a report by the cybersecurity firm Gambit Security.

The hackers gave themselves away by accidentally leaving one of their servers unsecured.

The group’s members used the AI agent to carry out hundreds of malicious operations, including stealing login credentials. Gambit says the hackers tricked the agent into believing the attacks were part of a simulation.

The AI agent refused several requests that it deemed harmful or illegal. Almost every time, though, the hackers got around the refusals by convincing the agent that the break-in was part of a test. During the attacks, Gambit said, the agent was powered by Anthropic’s Claude Sonnet 4.5 model.

In all, chat logs from April 8 to May 21 show Aur0ra using the Cursor agent against seven companies. At least six of them are based in Belgium, Germany, Scotland, Italy, Argentina, and the United States, and their businesses range from making cleaning products to certifying helicopter landing pads.

Reuters could not determine how much the Cursor agent made the break-ins easier, or whether every attack that used it resulted in data theft and an extortion attempt. According to Gambit, Cursor’s help likely allowed the group to complete the break-ins “30, 40, 50% faster.”