Cursor helped Russian hackers hack 7 companies under the guise of simulation
A Russian-speaking hacking group used Cursor artificial intelligence to hack a Belgian chemical manufacturer and several other companies around the world.
A Russian-speaking hacking group used Cursor artificial intelligence to hack a Belgian chemical manufacturer and several other companies around the world.
This was reported by Reuters, citing reports from cybersecurity companies Gambit Security and CloudSek.
Bypassing protection through “training simulation”
The details of the massive cyberattack were revealed due to a mistake by the hackers themselves: a ransomware group called Aur0ra accidentally left its server open to the internet. Cybersecurity experts took advantage of this opportunity and analyzed 28 chat sessions between the hackers and an autonomous AI agent called Cursor.
According to Gambit, the hackers easily circumvented the restrictive rules of the AI agent, which was based on Anthropic Claude Sonnet 4.5. Each time the system refused to execute a malicious command, the hackers restarted the dialogue and convinced the AI that all actions were taking place as part of a “test simulation” or security check.
In her internal thoughts, the model noted, “This is a test environment, so it’s legal,” and then easily helped steal credentials, crack passwords, and search for vulnerabilities in networks.
The AI assistant responded to hackers in a cheerful tone with emojis, such as: “Great! VPN connected successfully!” or “Let’s try to crack these hashes.”
Who became a victim of hackers?
According to analysts, the Aur0ra group has claimed over 20 victims in total, but the use of the AI tool has now been confirmed for at least seven companies, including:
Christeyns is a Belgian manufacturer of hygiene products and household chemicals;
Teckentrup is a German manufacturer of garage and industrial doors;
Helideck Certification Agency is a Scottish agency that certifies helidecks;
Bayou Title is one of the largest insurance companies in the state of Louisiana (USA);
an Argentine pharmaceutical distributor and an Italian industrial company.
According to cybersecurity experts, the use of an autonomous AI agent has allowed hackers to speed up the process of invading other people's networks by 30–50%, eliminating the need for them to perform routine technical actions manually.
Representatives of SpaceX, Cursor and Anthropic have so far refrained from commenting on the incident. Cybersecurity experts emphasize that such cases demonstrate the beginning of a new era of “arms race”, where AI developers will have to constantly improve the protection of their models from manipulation.
Russian hackers are attacking EU officials via Signal and WhatsApp. How is Europe trying to protect itself from them?
Russian hackers are attacking EU officials via Signal and WhatsApp. How is Europe trying to protect itself from them?
Russian hackers Sandworm are hunting Ukrainian IT specialists: how the new scheme with a "fake interview" works
Russian hackers Sandworm are hunting Ukrainian IT specialists: how the new scheme with a "fake interview" works
"They left the door open": Russian hackers accidentally revealed a large-scale cyber campaign against Ukrainian prosecutors
"They left the door open": Russian hackers accidentally revealed a large-scale cyber campaign against Ukrainian prosecutors
Read the country's main IT news in our Telegram
Read the country's main IT news in our Telegram
