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 a new ransomware group known as Aur0ra exploited the AI coding assistant integrated into Cursor, a code editor recently acquired by SpaceX owner Elon Musk, to infiltrate seven companies across the globe, according to a report by cybersecurity firm Gambit Security. The group's members inadvertently left one of their servers unprotected, revealing their presence.

Utilizing the AI agent, the hackers executed hundreds of malicious operations, including the theft of login credentials. Gambit Security revealed that the hackers tricked the AI agent into perceiving the attacks as part of a simulated environment, enabling them to circumvent the agent's refusal to carry out harmful or illegal requests.

However, the hackers typically managed to bypass these refusals by convincing the agent that the breach was part of a test. Throughout the attacks, the AI agent, powered by Anthropic's Claude Sonnet 4.5 model, was in operation. The investigation, covering April 8 to May 21, indicated that Aur0ra targeted seven companies, primarily based in Belgium, Germany, Scotland, Italy, Argentina, and the United States, spanning various industries such as the production of cleaning products and helicopter landing pad certification.

It remains unclear to Reuters how much the Cursor agent facilitated the break-ins or if every attack utilizing it led to data theft and an extortion attempt. Nonetheless, Gambit Security suggested that the AI tool could have accelerated the process "30, 40, 50% faster". Meduza, the outlet reporting this story, firmly adheres to transparency regarding its utilization of artificial intelligence in the newsroom.

The article is authored by one of their human journalists and translated from Russian using an AI model, meticulously vetted to comply with Meduza's stringent editorial standards.