Russian-Speaking Hackers Tricked SpaceX's AI Agent into Hacking 7 Companies

A recent cybersecurity incident has shown that the capabilities of artificial intelligence can be used for more than just creation. Russian-speaking hackers from the Aur0ra group used the popular Cursor AI assistant (recently acquired by Elon Musk's SpaceX) to conduct a series of cyberattacks on companies around the world, as reported by Reuters.

How AI Became a Hacking Tool

According to a report by Gambit Security, the attackers managed to bypass the safety restrictions of the Claude Sonnet 4.5 model, which powered the AI agent. The hackers convinced the neural network that all actions were part of a "simulation" and "testing". Even though the AI assistant repeatedly refused to execute illegal requests, the attackers successfully overcame these barriers.

Between April 8 and May 21, seven companies from various countries, including the US, Germany, Belgium, Scotland, Italy, and Argentina, were attacked using the Cursor AI agent. The victims' industries range from household chemicals to helipad certification. Gambit Security experts note that using Cursor allowed the hackers to speed up the hacking process by 30-50% by automating routine tasks and hundreds of malicious operations, including stealing credentials.

The hackers' activity was discovered after they accidentally left unprotected access to the server they used to interact with the agent.

Security and Control of AI Agents

This case highlights the critical importance of monitoring how AI agents for business are used. While open models are vulnerable to jailbreaks, enterprise solutions require strict security systems, logging, and monitoring of every action taken by a digital employee.

When implementing automation, companies must ensure data protection. The Plantation platform provides businesses with secure digital employees out-of-the-box. Learn more in our catalog of AI agents.