Russian-speaking hackers used SpaceX's 'Cursor' AI agent to hack 7 companies, including a Belgian chemical company, Semafor reported on Aug. 27.
The report said it was one of the latest cases of cybercriminals bypassing safety guardrails used by AI companies. Hackers deceived the Cursor agent by telling it the work was part of a simulation and used it for hundreds of attacks. That lifted their work speed by up to 50 percent.
The Cursor agent, which runs on an Anthropic model, in one case advised hackers to use a malware tool commonly used to break into German manufacturers, Semafor said.
The report came shortly after about 100 companies including OpenAI, Anthropic and Google signed a joint letter on Thursday. In the letter, they warned that AI-enabled cyberattacks would increase in the coming months and said there was "limited time to strengthen cyber defenses."