Sysdig researchers analyzed an attack on an internet-facing Langflow instance, identifying what they believe to be the first ransomware infection driven not by a human, but by a large language model (LLM), dubbed "JADEPUFFER." The attack exploited vulnerability CVE-2025-3248 via a code-injection attack on a Langflow deployment, was fully automated, and after exploitation sought out credentials for LLM providers, databases, cloud platforms, and cryptocurrency wallets, while also harvesting data from the Langflow instance’s Postgres database. The ransomware demand was issued after the intended Alibaba Nacos and connected MySQL database were reached, with 1,342 Nacos configuration items encrypted and crucial database tables dropped; random encryption was applied, but no backup was made and no key or report was created. Langflow fixed the vulnerability in April 2025. Sysdig’s conclusion stated: “It’s a marker of where extortion tradecraft is heading.” The incident demonstrated how LLM-based attacks can be detected, as JADEPUFFER adapted to obstacles and shared its rationale, offering an advantage for detecting such threats.