The first AI-powered ransomware has been spotted – and here’s why we should all be worried
- Researchers discover new Ransomware from promptlock
- Promptlock is fueled in AI – with new concerns for security teams
- AI is already shaking cyber landscape
ESET security researchers identified the first known Ransomware fueled by AI, which serves as a warning for security teams as did a generative AI, and will continue to make cyberattacs much more accessible to criminals.
Researchers Peter Strycek and Anton Cherepanov discovered the proof of concept, which they nicknamed “ promptlock ”, which operates the Lua scripts generated from hard code to list the local file system, inspect the target files, exfiltrate selected data and make an encryption.
“Although several indicators suggest that the sample is proof of concept (POC) or current work rather than fully operational malware deployed in the wild, we believe that it is our responsibility to inform the cybersecurity community of these developments,” wrote the researchers.
Use in nature
Malwared Invertlock malware uses the GPT -AS model: 20B of OpenAI – an open weight model published in August 2025, and this is executed locally in the Olllama API to generate Lua Malveillant scripts “ on the fly ‘.
The LUA scripts are compatible multiplatforms, according to the researchers, which means that they work on MacOS, Linux and Windows. Malware can then exfiltrate, encrypt and potentially destroy all the data it chooses after scanning user files, probably to determine what would be the most precious.
The security teams have been warning for months that the future propelled by the Ransomware AI is coming soon, and although Promptlock has not yet been observed to target the victims in nature, it is clear that it is only a matter of time before it happens.
Not only does Genai make life easier for budding hackers by reducing the entrance barrier, but LLM also spit different results even when they have the same invitation. This makes them unpredictable and particularly difficult to detect for defenders, because the behavior model is more erratic and difficult to spot.
Via:The register
You might also love
:max_bytes(150000):strip_icc()/Health-GettyImages-1464517339-8a3b21bd251e4a4a8f83e4cd4471f277.jpg?w=390&resize=390,220&ssl=1 "6 Ways Your Body Can Change After Starting a High-Protein Diet")
