The Age of Ransomware Created by AI Has Begun

While these occurrences haven’t become standard in the ransomware landscape, the insights serve as a significant alert.

“There are some groups leveraging AI in the evolution of ransomware and malware, but from what Recorded Future can see, most aren’t,” states Allan Liska, a Recorded Future analyst specializing in ransomware. “The broader usage of AI is observed primarily in the realm of initial access.”

In a related development, researchers at cybersecurity firm ESET recently revealed the discovery of the “first known AI-powered ransomware,” named PromptLock. This malware predominantly operates locally on a device and utilizes an open-source AI model from OpenAI, enabling it to “generate malicious Lua scripts on the fly.” It employs these scripts to inspect targeted files, steal data, and initiate encryption. ESET theorizes that the code serves as a proof-of-concept, likely not yet deployed against any victims, but stresses that it highlights how cybercriminals are beginning to integrate LLMs into their toolkit.

“Implementing AI-assisted ransomware poses specific challenges, mainly due to the substantial size of AI models and their elevated computational needs. Nonetheless, cybercriminals may discover methods to overcome these obstacles,” wrote ESET malware researchers Anton Cherepanov and Peter Strycek, who identified the new ransomware, in an email to WIRED. “Regarding development, it is highly probable that threat actors are actively pursuing advancements in this area, and we will likely witness more attempts to engineer increasingly complex threats.”

Though PromptLock has yet to be utilized in practice, findings from Anthropic further emphasize the rapidity with which cybercriminals are incorporating LLMs into their operations. The AI firm also identified another cybercrime group, labeled GTG-2002, employing Claude Code for automatic target identification, gaining entry into victim networks, crafting malware, and subsequently exfiltrating data, analyzing the stolen material, and generating ransom notes.

In the past month, this attack has affected “at least” 17 entities across government, healthcare, emergency services, and religious sectors, according to Anthropic, which refrains from disclosing the names of the impacted organizations. “This operation indicates a troubling evolution in AI-assisted cybercrime,” noted Anthropic’s researchers in their report, “where AI operates as both a technical advisor and active participant, facilitating attacks that would otherwise be more challenging and time-consuming for individual actors to carry out manually.”