First AI Virus Discovered
Credit: Pixabay
Analysts at the Slovak company ESET Research Labs have recorded the emergence of a qualitatively new type of cyber threat. The malicious PromptLock has become the first program to use machine learning capabilities to create attack code on the fly.
According to the researchers, PromptLock uses a local implementation of the gpt-oss:20b model via the Ollama API, which allows it to work in Windows, macOS, and Linux environments without the need to create separate versions. Such capabilities increase the flexibility of the virus and make it difficult to detect. PromptLock’s code is written in Go and is classified as Filecoder.PromptLock.A. The malware scans the file system and then selectively exfiltrates data and encrypts it using the 128-bit Speck algorithm. PromptLock’s code itself does not contain a language model; it connects to it through its own server via a proxy, which allows it to bypass network restrictions.
ESET experts emphasize the importance of proactively informing the community about such developments. Companies using AI technologies must prepare to counter a fundamentally new class of threats, where machine learning becomes a weapon in the hands of cybercriminals.
