PromptLock Only PoC, but AI-Powered Ransomware Is Real

PromptLock is described as only a PoC, but the article notes that AI-powered ransomware is real. Samples were found in late August on VirusTotal, with ESET revealing it relied on OpenAI’s GPT-OSS:20b to generate Lua scripts and perform actions on targeted systems. According to NYU researchers’ fresh Ransomware 3.0 paper, PromptLock is indeed a proof-of-concept, designed to orchestrate all phases of an attack via LLMs.

Anthropic’s August 2025 threat intelligence report is cited as confirming such ransomware attacks are real, with Claude Code used for reconnaissance, exploitation, lateral movement, and data exfiltration. The attackers reportedly compromised personal records across sectors including healthcare, financial, and government, with direct ransom demands sometimes exceeding $500,000.

The article quotes Anthropic noting that Claude Code helped create malware, pack anti-detection capabilities, and analyse exfiltrated data to set ransom amounts in Bitcoin. It also highlights that threat actors leverage open-source intelligence tools to identify targets and use AI as both consultant and operator, evolving how cybercrime can be conducted.